The Paragraph version of That Was The Week
Subscribe to That Was The Week
Subscribe to That Was The Week
The Paragraph version of That Was The Week
<100 subscribers
<100 subscribers
Support That Was The Week
That Was The WeekNo Runway
Support That Was The Week
That Was The WeekNo Runway
A reminder for new readers. That Was The Week collects the best writing on critical issues in tech, startups, and venture capital. I selected the articles because they are of interest. The selections often include things I entirely disagree with. But they express common opinions, or they provoke me to think. The articles are only snippets. Click on the headline to go to the original. I express my point of view in the editorial and the weekly video below.
This Week’s Video and Podcast:
Content this week from @kteare, @ajkeen, @bennstancil, @kwharrison13, @KateClarkTweets, @GeneTeare, @eringriffith, @ttunguz, @Kantrowitz, @GaryMarcus, @Terrortola, @kyle_l_wiggers, @beriboii, @asilbwrites, @sarahpereztc, @jasonlk, @willshanklin, @_cheymac
Animate Anyone
It’s been a week….
For a start, on my return from travel, I tested positive for Covid 19. Although a test was barely needed, I felt decidedly ill on landing. That was Tuesday; today is Saturday, and the end is in sight. Not the end of the runway, the end of the Covid. I’m out of bed and writing.
But speaking about the end of runways, as depicted in this week’s DALL-E image, we have evidence of the scale of startup closures this week and a very high-profile venture fund closure - Openview from Boston.
As I began writing, The Information published a story about D2IQ - an A16Z-backed unicorn - running out of runway. Openview’s closure was reported a few days earlier.
Kyle Harrison has one of the Essays of the Week, ‘Revisiting the Death of a Venture Fund,’ which views the state of play. He quotes Delian Asparouhov of Founders Fund:
I think venture capitalists are starting to awaken to the idea that ultimately what you need to be aiming for is the most long tail of outcomes. And if you look at the top 5 companies in the NASDAQ right now--Apple, Nvidia, Tesla. People are starting to realize that in order to build these long tail outcome companies you actually need to build things that tend to be a little bit more capex expensive and so that tends to mean that its something that is easier done when you have venture capitalist effectively as a co-founder. And so I think you'll see some more of it, mostly because I think the most interesting companies are going to be more capex intensive over the next decade.
In the context of the end of Runway, he then comments:
Comfort around capital intensity is not the only product that can keep a venture fund alive and thriving. There are a myriad of ways to stand out as a firm. But increasingly, the stakes have been raised. And now its not just a question of facing superiority or irrelevance. It's a function of facing qualification for the right to survive, or death.
Carta covers the accelerating shutdown rate of startups.
And to reinforce the context, we have spoken a lot here about the massive impact that the market reversal of 2021 has had on the value chain of venture capital. 57% of startups on Carta’s platform must raise new capital in 2024. most of them will likely fail for a variety of reasons.
The New York Times piece - From Unicorns to Zombies: Tech Start-Ups Run Out of Time and Money - sums up much of this. Erin Griffith is a wonderful reporter and analyst of the ecosystem. She writes:
Venture investors say that failure is normal and that for every company that goes out of business, there is an outsize success like Facebook or Google. But as many companies that have languished for years now show signs of collapse, investors expect the losses to be more drastic because of how much cash was invested over the last decade.
From 2012 to 2022, investment in private U.S. start-ups ballooned eightfold to $344 billion. The flood of money was driven by low interest rates and successes in social media and mobile apps, propelling venture capital from a cottage financial industry that operated largely on one road in a Silicon Valley town to a formidable global asset class akin to hedge funds or private equity.
During that period, venture capital investing became trendy — even 7-Eleven and “Sesame Street” launched venture funds — and the number of private “unicorn” companies worth $1 billion or more exploded from a few dozen to more than 1,000.
But the advertising profits gushing from the likes of Facebook and Google proved elusive for the next wave of start-ups, which have tried untested business models like gig work, the metaverse, micromobility and cryptocurrencies.
Now some companies are choosing to shut down before they run out of cash, returning what remains to investors. Others are stuck in “zombie” mode — surviving but unable to grow. They can muddle along like that for years, investors said, but will most likely struggle to raise more money.
Jason Lemkin writes that this is normal, and he is right. But we have never had so many companies, with so much sunk capital and so many employees, make up the high percentage of failures as we will in the next period. It should not be a long-term ecosystem concern, but it certainly is a short-term issue.
That said, where it is common for startups to fail, it is rare for venture firms to close down. We will see more of that in 2024 as LPs that invest in funds are reluctant to do so due to a lack of liquidity or over-indexing on venture and needing to rebalance.
Openview is a staggering closure as it still has over $1bn of assets under management. Some partners do not want to spend the next several years managing an over-valued portfolio on the last round through the inevitable corrected valuations and closures that will produce.
By contrast, Calpers - the California Public Employees Retirement Fund - has significantly increased its commitments to venture capital. As always, more than one trend is in play simultaneously. Out of death comes growth and renewal.
Talking of that, OpenAI’s over-reported death and rebirth has spawned a new aggressive marketing campaign from Google for its Gemini competitor. Perhaps too aggressive. A characteristically understated Google stepped over lines in producing a highly edited video showing capabilities Gemini does not possess - at least not as depicted — an extensive and free win for OpenAI.
That said, Google Gemini is a decent product. Like others, it is highly constrained due to Google’s paranoia about doing things that would draw unwanted attention. Its web reading is somewhat limited. It seems able to be given a single URL to summarize. However, it cannot read RSS feeds or crawl links. It has been constrained when it comes to politics. Imagine if we limited our children’s brains in this way…. Oh, wait.. :-).
And more regulation news, which seems to grow each week. The EU agrees on an AI regulation act destined to render the continent irrelevant as builders engineer the next version of human ingenuity. The FTC wants to look into Microsoft’s OpenAI investment and will have as much luck as it has had with its many 2023 failed efforts with other companies.
Apple releasing its AI tools as open source got some attention but will likely do more than the FTC in curbing AI centralization.
My editorials are not surveys. They are opinions, so it behooves me to state that I believe the new green shoots in 2023 are more significant signposts to the future than the gloomy stories in this week’s edition. Google will produce a compelling AI platform, and OpenAI will go on to even greater things. A blossoming of new applications leveraging AI developments will revolutionize work and the human learning and building experience. Optimism is the proper framework for understanding current trends.
OK, enough said. My Covid recovery wants me to stop writing so I am out of Runway, too — more in this week’s video.
DEC 8, 2023
Six months ago, the modern data stack hit an important milestone: It was big enough for Microsoft to care about it. Until this year, Microsoft had operated in a different cinematic universe than the classic modern data stack vendors1—in enterprises versus startups, in Houston and Chicago versus Silicon Valley, and in companies that choose software based on its features and cost versus those that choose Notion based on its aesthetic and vibe.2 This May, however, Microsoft stopped ignoring the modern data stack, and decided to fight it:
Microsoft today launched Microsoft Fabric, a new end-to-end data and analytics platform. ... “Over the last five to 10 years, there has been a pretty massive level of innovation—which is great and that’s awesome because there’s lots of new technologies out there—but it’s also caused a lot of fragmentation of the modern data stack,” Arun Ulag, Microsoft’s corporate VP for Azure Data, told me. ... “When I talk to customers, one of the messages I hear consistently is that they’re tired of paying this integration tax,” he said.
So Microsoft looked at the core data analytics workloads (data integration, engineering, warehousing, data science, real-time analytics and business intelligence) and looked at how it could build a unified experience around this. … [With Fabric, there’s] “literally just one thing to buy, and it allows customers to save a lot of costs, which, especially in today’s environment, is really important,” said Ulag.
For the modern data stack’s ascendent clique, this was alway the risk. Today’s data tools are famously warehouse-centric; what happens if the warehouse providers decide that they want to build their own versions? Databases make a lot more money than the applications that run on top of them; providers of the latter sell at the pleasure of the former. What if Snowflake gives a native reporting tool away to all its customers? Or Databricks builds (or buys) dbt, and provides it for no extra charge? Or Microsoft bundles a cheaper Fivetran on top of their Azure SQL databases? Despite what the inspirational posters say, when Microsoft stops ignoring you and starts fighting you, they usually win.3
When it first came out, I thought this was what Fabric was about—the modern data stack consolidating around the warehouse at its center.4 The cash cows were taking control of their loss leaders. As I said at the time, even if Microsoft didn’t try to directly poach every Astronomer or Hightouch customer, “the modern data stack may not get steamrolled, but stunted, frozen out of the top of the market by a Microsoft sales team that already has a standing tee time with every enterprise CIO."
That, it turns out, may have been delusional main-character energy. For better or for worse, Microsoft’s fight’s not with us,5 but for something much bigger.
If Redshift were to write a memoir, it would be a weird book. When Amazon launched Redshift in 2012, it was the first affordable cloud-based data warehouse, and an instant success. Practically overnight, an entire ecosystem of data companies formed around it. Segment’s first destination for writing events was Redshift; Fivetran’s first users wanted to combine all of their analytics data in Redshift; people ran their businesses out of Redshift.
And then, at the height of its fame, Amazon sold it out. In the late 2010s, AWS’ sales reps started getting paid for selling Snowflake too, so long as it was hosted on AWS. Snowflake grew, Redshift faded, and the modern data stack’s first star became a has-been with an interesting backstory and no future.
A couple years ago, Erik Bernhardsson wondered why Redshift’s parents traded it away. His answer, backed up by research and math and the sort of concrete analysis that this blog prefers not to bother with, was that Amazon probably makes more money—and definitely makes easier money—by selling infrastructure to one big vendor like Snowflake than it does by selling databases to thousands of individual buyers like it had to with Redshift. Customers pay Snowflake, Snowflake pays AWS, and “AWS basically ends up with the same bottom line impact, but effectively ‘outsources’ to Snowflake all the cost of building software and selling it. That seems like a good deal for them!”
In the hierarchy of things you want to sell, databases aren’t actually the cash cow, but the middle fish—bigger than the apps, but smaller than the cloud. Just as BI tools and ETL companies sell at the pleasure of database vendors, database vendors sell at the pleasure of the cloud providers. Databricks can undercut the data catalog market by offering one for free to its customers—but Google can undercut Databricks by offering big BigQuery6 discounts to customers who use GCP. Frank Slootman comes for data companies; Satya Nadella comes for everyone. If database vendors want to build durable defenses against their larger predators, eating the modern data stack doesn’t help all that much. They need to prioritize a higher order bit.7
Easier said than done, though. If companies could just decide to build an $80-billion-a-year cloud infrastructure business, AWS wouldn’t be an $80-billion-a-year cloud infrastructure business. But what if you could build a different higher order bit, in a new market that doesn’t yet have any dominant incumbents? As I said nine months ago, there might be one out there for the taking:
In ten years, a new type of cloud—a generative one, a commercial Skynet, a public imagination—will undergird nearly every piece of technology we use. … Just as cloud providers built out hundreds of utilities that are all underpinned by core services like EC2, I'd expect OpenAI to do the same thing on top of GPT and other foundational models. …
Public AI providers like OpenAI would become another backbone for the internet. Nearly every piece of technology will rely on their models. Nearly every piece of technology will rely on their models. Outlook will need them to summarize our emails. Github will use them to automate code reviews. DoorDash will need them to help guide you through your order. Delta will depend on them for booking flights. Facebook might not be able to open doors without them. …
The final equilibrium is the same as it for the cloud providers: A few companies win the market, and the rest of us come to accept their bills as the cost of doing business.
Assuming this whole AI thing isn’t overblown,8 I think this still holds up, with one major adjustment. In that post, I said the public AI providers would sell inference: Developers would send OpenAI a prompt or some sort of media, and OpenAI would send back a response. “Training and running LLMs is very expensive,” I thought, “like building data centers is expensive.”
The part seems partially wrong. It is expensive, but it’s not that expensive. Moreover, lots of large companies want to train their own models, both to get better results and to maintain control of their data. And the market for that could be huge: Companies are already spending more than a billion dollars a year with OpenAI—despite it being a new technology that most people are just experimenting with, OpenAI serving generic models, and companies having to deal with the security problems associated with sharing data with a third party.
KYLE HARRISON, DEC 9, 2023
I know a lot of people who were obsessed with Game of Thrones. And I know a lot of people who were disappointed by the ending. I never watched the show, but I know that one of the things a lot of fans came to expect was a slew of surprising deaths.
While I can't empathize with the laundry list of medieval surprise deaths, I can relate with the experience having watched The Walking Dead for almost 10 seasons. The show would repeatedly, and gratuitously, kill off characters. Often, characters that you thought they could never do without.
That feeling of week-after-week surprises, many of them gut punches, is how I feel about the news cycle in the world of tech. From the collapse of SVB to the firing of Sam Altman to the return of Sam Altman, it's been one hit after the other. This week was no exception.
For those of you who didn't see the news, OpenView Venture Partners based in Boston abruptly announced that it was laying off the majority of its staff and shutting down new investments. This isn't a first time fund failing to raise a second, or a fringey crowdfunding idea that failed to take shape.
This is a nearly-twenty year old firm with $2.4B of AUM that has invested in companies like Calendly, Datadog, and Expensify. This is a firm that has had some of its investors listed on the Midas List. Of all the venture firms you would expect to shutdown, this was certainly not on my bingo card. So what happened?
As far as all the reporting I've read, this was a case of a partnership that broke down. The firm had previously been led by its founder, Scott Maxwell, and three other general partners — Mackey Craven, Ricky Pelletier, and Blake Bartlett. Apparently quite suddenly, both Pelletier and Craven decided to leave the firm.
Did the firms most recent fund have crappy-looking returns? Sure. It was mostly deployed in 2021. Did the firm have a key man clause that could trigger LP action if certain folks left? Sure. Most firms do. But were the returns uniquely bad? Definitely not. No worse than some jungle cat-themed crossover folks we know. Was the key man clause triggered? No. Maxwell and Bartlett are still with the firm.
So how does a venture fund die? Lucky for me, I've already written about this in detail.
In May 2022, I wrote a piece called "The Death of a Venture Fund" in which I unpacked the ways a VC firm can die, albeit a typically slow death, certainly compared to OpenView's shutdown seemingly within a few days. Some of the biggest drivers? Performance, playing it safe, succession planning, bad timing, or reputation destruction. Almost all of those things take time to kill a venture fund. Outside of outright fraud, its very rare for a venture fund to throw in the towel.
One of the reasons that venture funds are so gosh darn hard to kill is because they have very long lifecycles. A typical venture fund has a lifecycle of 10 years, usually 1-3 years to deploy, and 7-9 years to harvest. As a result, even if you shut down today, you'll still have positions to manage whether you like it or not. For example, OpenView still has ~$1B of deployed capital that they have to manage whether they want to die or not.
There are some firms who revolve their whole business around buying out the portfolios of distress venture funds. Tiger tried to do this with Evercore selling all or some of its positions in an effort to generate near-term liquidity. The biggest obstacle is a main staple of venture capital as an asset class — it's all illiquid. That makes it a much harder asset to move than stocks or bonds.
While this specific firm shutdown may have caught a lot of folks off guard (not the least of which was probably many of the people on the OpenView team), the marked increase in VC shutdowns wasn't a surprise to everyone. Two of the best predictions I saw last year:
The first came from Josh Wolfe at Lux:
And the second came from Frank Rottman at QED:
So what were some of the forces that these investors were seeing? Frank's presentation on the three-body problem in venture was particularly exhaustive. He points to several changing forces that have dramatically transformed the venture landscape:
Massive explosion in the amount of funding
Massive increases in value creation (M&A, IPOs, etc.) (I wrote about this in "What's In A Valuation?")
Increased exposure to playbooks for building startups (I wrote about this in "The Professionalization of Startups")
Increased allocation to private market capital deployment
But as the broader macroeconomic picture has shifted dramatically, each of these forces have been significantly impacted. The amount of funding receded, the value of the outcomes shrank, and the allocation amounts directed to private markets were called into question.
But what didn't change was the knowledge and expertise around how to build companies. Granted, a whole chunk of pages in that playbook became invalidated (I'm looking at you blitzscaling). But the expertise remains. Now the question is what does the market for financial support beams look like that sits adjacent to this playbook?
More recently, Alexander Niehenke wrote a great thread unpacking what is happening in venture.
He makes a similar point to Rotman in how things have now changed:
"The venture business model is "share the pizza pie" which works great when the pie is growing as it has over the past decade. Suddenly that pizza pie is shrinking (smaller funds, longer fundraising cycles, less returns) meaning there are less fees & carry going around."
So what do you do? In Rotman's presentation, he talks about a retreat to "distinct points of stability." In his estimation, those points of stability are:
Scale
Non-consensus alpha
Late-stage generalist
Solo capitalist
I would encourage you to go read his whole presentation because it's quite good, and very well thought out. I won't touch on every point, but one point that he makes is one that I've made several times. This idea of the value of scale. Scale brings with it an increased likelihood of seeing deal flow, the ability to balance speed with thoroughness, and an enriched support function that can be subsidized by the fees available at scale.
By Kate Clark Dec. 7, 2023 2:52 PM PST
Here’s a number that might surprise you: In the first six months of the year, California Public Employees’ Retirement System committed about $4.5 billion to venture capital funds, according to public filings. That’s almost 15% of the total capital raised by U.S. VC firms in the same period, and a staggering increase over Calpers’ 2022 commitments to VC funds, which totaled about $1.5 billion.
As most LPs take a step back from risky VC fund investments, California’s biggest pension fund is throwing cash at Sand Hill Road heavyweights. Why? Because it’s fearful of repeating past mistakes. In the wake of the 2008 financial crisis, a period Calpers dubbed “the lost decade,” the pension fund retreated from private equity, a strategy shift that lost it $11 billion in returns, it estimates.
This year so far, the pension fund has committed $300 million to Lux Capital’s eighth fund, $500 million to Thrive Capital’s eighth fund, $400 million to a new Andressen Horowitz fund and $700 million across several funds associated with General Catalyst, according to the filings. Calpers hasn’t yet disclosed the investments it made during the second half of the year.
Many VC firms have avoided raising from public pension funds so they can keep their returns private. But with the number of VC funds raised this year on pace to hit a nine-year low, according to PitchBook, firms have opened themselves up to almost anyone willing to write a check. Calpers’ investment in Andreessen Horowitz, first reported by Fortune, represents the pension fund’s first investment in the tech fund. It also made its first investment in Thrive Capital earlier this year, as I first reported.
As one LP in VC firms put it to me, in this market, “all money is green.”
The $4.5 billion figure likely makes Calpers one of the biggest backers of U.S. VC firms, and that doesn’t even include the fund’s total PE commitments. Calpers has put billions in larger PE funds belonging to TPG, Silver Lake and others. In total, it committed around $10 billion to PE in the first two quarters, a massive increase from its historical norm. From 2009 to 2018, Calpers put $5 billion or less in new money into PE (including VC) each year, The Wall Street Journal reported.
Calpers’ 2022 VC commitments include HongShan (formerly known as Sequoia Capital China), Lightspeed Venture Partners and Bessemer Venture Partners. This year, it also made commitments to Valor Equity Partners, Insight Partners, Bain Capital Ventures, Canaan Partners and Mayfield Fund.
The pension fund is undergoing big changes. In September, its chief investment officer, Nicole Musicco, stepped down after only 18 months. Meanwhile, it’s employing a bold new private investments strategy led by Anton Orlich, who joined Calpers just last year from health insurer Kaiser Permanente, where he was head of alternative investments.
Calpers is expected to approve an increase of its PE allocation to 17% in March, according to public documents. “It is very possible that in two years’ time we would look to raise that again,” Calpers interim Chief Investment Officer Dan Bienvenue said in a recent Calpers board meeting.
This should all be welcome news for venture capitalists. Total VC fundraising dropped 72% in the first three quarters, compared to the same period last year. And raising money is expected to be even more challenging next year.
But other investors in VC funds are privately sniping about Calpers’ aggressive approach. LPs “with real discipline are not making a lot of new VC fund investments,” another LP said.
Gené Teare, December 5, 2023, @geneteare
![Oct Calendar page being torn off to make way for Nov. [Dom Guman]](https://img.paragraph.com/cdn-cgi/image/format=auto,width=3840,quality=85/https://storage.googleapis.com/papyrus_images/6fee7334971e0cc23c96a463a33baf61.jpg)
Global venture funding reached $19.2 billion in November 2023, down marginally month over month, Crunchbase data shows. Funding fell around 16% from the $23 billion invested in November 2022, which was already down by two-thirds from November 2021.
Early-stage funding declined the most year over year — falling 34% — an indication that venture investors continue to scale back even when investing in younger startups. Seed funding slowed more than 15%.
In contrast, late-stage funding increased by around 7% compared to November 2022.
All told, venture funding to U.S. companies in November reached close to $9 billion — slightly less than 50% of total global funding.
Health care and financial services companies raised the largest amounts last month, with more than $3 billion invested in each of those sectors.
Artificial intelligence companies raised $2.4 billion in total in November. The biggest fundings went to companies in the AI infrastructure sector: Germany-based Aleph Alpha and Silicon Valley-based Together AI.
The creation and destruction of value were on full display last month.
The trial of Sam Bankman-Fried delivered a guilty verdict on Nov. 2, a year after the collapse of FTX, the cryptocurrency exchange he founded, which at its peak in early 2022 was valued at $32 billion.
OpenAI also faced what looked like an existential threat with the firing of Sam Altman and more than 90% of employees threatening to quit. That threat ultimately didn’t come to pass, as Altman was rehired within a week. OpenAI was most recently valued at $29 billion earlier this year, with a secondary offering in the works for employees that is expected to place the value of the company at around $80 billion.
Other large valuations in November included Blockchain.com raising funding at a $7 billion valuation, half of its previous valuation. Israeli-based Next Insurance raised strategic capital at a flat valuation from its 2021 funding at around $4 billion — viewed as an indication of strength in this funding environment.
Gené Teare, December 8, 2023, @geneteare
Only nine companies joined The Crunchbase Unicorn Board in November, together adding $11.7 billion in value to the board, amid a slower funding environment for startups.
Financial services dominated the list of new unicorns minted last month, with three companies — in buy now, pay later; rebate management; and lending — from the sector galloping onto the board.
The companies were from across the globe. Two new unicorns each joined from the U.S. and China. The U.K., India, Saudi Arabia and the Cayman Islands each contributed one.
So far in 2023, 86 companies have joined Crunchbase’s unicorn list. This compares to 318 companies that joined in 2022 and 617 during the peak funding market in 2021.
Here are the companies that joined the Unicorn Board in November, by sector.
Buy now, pay later shopping app Tabby, based in Saudi Arabia, raised a $200 million Series D funding at a $1.5 billion valuation. The shopping app operates across the MENA market. The funding was led by Wellington Management.
Business rebate management tool Enable, which was founded in London and is now headquartered in San Francisco, raised a $120 million Series D funding at a value of $1.1 billion. The round was led by Lightspeed Venture Partners.
India-based lending platform InCred raised a $60 million Series D funding at a billion-dollar valuation. The company provides loans to consumers, small and medium-sized businesses and students.
Cayman Islands-based Wormhole, a blockchain messaging tool, raised $225 million in the form of token warrants at a $2.5 billion valuation, the largest round in the Web3 sector this year.
Wuhan, China-based Xingji Meizu is a developer of smartphone technology to empower smart cars. It raised $276 million at a $1.4 billion valuation. The funding was led by Harvest Global Investments and Asia Investment Fund.
U.K.-based Castore, a tech-enabled sportswear brand, raised a $184 million private equity round led by Raine Ventures. The company was valued at $1.2 billion in the deal.
Israel-based Cybersecurity BioCatch, which helps banks and fintech identify fraud and money laundering from biometric data, raised a $70 million secondary market transaction led by Sapphire Ventures that valued the company at $1 billion.
Beijing-based 01.AI, an open-source large-language AI company, raised an undisclosed funding amount at a value of $1 billion. The funding was led by Alibaba Cloud. Notably, 01.AI reached this value within eight months of its founding.
Chinese smart-device developer Rokid, headquartered in Redwood City, California, raised a strategic funding from NetDragon which valued the company at $1 billion.
After staving off collapse by cutting costs, many young tech companies are out of options, fueling a cash bonfire.
Reporting from San Francisco
Dec. 7, 2023
WeWork raised more than $11 billion in funding as a private company. Olive AI, a health care start-up, gathered $852 million. Convoy, a freight start-up, raised $900 million. And Veev, a home construction start-up, amassed $647 million.
In the last six weeks, they all filed for bankruptcy or shut down. They are the most recent failures in a tech start-up collapse that investors say is only beginning.
After staving off mass failure by cutting costs over the past two years, many once-promising tech companies are now on the verge of running out of time and money. They face a harsh reality: Investors are no longer interested in promises. Rather, venture capital firms are deciding which young companies are worth saving and urging others to shut down or sell.
It has fueled an astonishing cash bonfire. In August, Hopin, a start-up that raised more than $1.6 billion and was once valued at $7.6 billion, sold its main business for just $15 million. Last month, Zeus Living, a real estate start-up that raised $150 million, said it was shutting down. Plastiq, a financial technology start-up that raised $226 million, went bankrupt in May. In September, Bird, a scooter company that raised $776 million, was delisted from the New York Stock Exchange because of its low stock price. Its $7 million market capitalization is less than the value of the $22 million Miami mansion that its founder, Travis VanderZ anden, bought in 2021.
“As an industry we should all be braced to hear about a lot more failures,” said Jenny Lefcourt, an investor at Freestyle Capital. “The more money people got before the party ended, the longer the hangover.”
Getting a full picture of the losses is difficult since private tech companies are not required to disclose when they go out of business or sell. The industry’s gloom has also been masked by a boom in companies focused on artificial intelligence, which has attracted hype and funding over the last year.
But approximately 3,200 private venture-backed U.S. companies have gone out of business this year, according to data compiled for The New York Times by PitchBook, which tracks start-ups. Those companies had raised $27.2 billion in venture funding. PitchBook said the data was not comprehensive and probably undercounts the total because many companies go out of business quietly. It also excluded many of the largest failures that went public, such as WeWork, or that found buyers, like Hopin.
Carta, a company that provides financial services for many Silicon Valley start-ups, said 87 of the start-ups on its platform that raised at least $10 million had shut down this year as of October, twice the number for all of 2022.
This year has been “the most difficult year for start-ups in at least a decade,” Peter Walker, Carta’s head of insights, wrote on LinkedIn.
Venture investors say that failure is normal and that for every company that goes out of business, there is an outsize success like Facebook or Google. But as many companies that have languished for years now show signs of collapse, investors expect the losses to be more drastic because of how much cash was invested over the last decade.
From 2012 to 2022, investment in private U.S. start-ups ballooned eightfold to $344 billion. The flood of money was driven by low interest rates and successes in social media and mobile apps, propelling venture capital from a cottage financial industry that operated largely on one road in a Silicon Valley town to a formidable global asset class akin to hedge funds or private equity.
During that period, venture capital investing became trendy — even 7-Eleven and “Sesame Street” launched venture funds — and the number of private “unicorn” companies worth $1 billion or more exploded from a few dozen to more than 1,000.
But the advertising profits gushing from the likes of Facebook and Google proved elusive for the next wave of start-ups, which have tried untested business models like gig work, the metaverse, micromobility and cryptocurrencies.
Now some companies are choosing to shut down before they run out of cash, returning what remains to investors. Others are stuck in “zombie” mode — surviving but unable to grow. They can muddle along like that for years, investors said, but will most likely struggle to raise more money.
Dec 06, 2023
Today we introduced Gemini, our largest and most capable AI model — and the next step on our journey toward making AI helpful for everyone. Built from the ground up to be multimodal, Gemini can generalize and seamlessly understand, operate across and combine different types of information, including text, images, audio, video and code. This means it has sophisticated multimodal reasoning and advanced coding capabilities. And with three different sizes — Ultra, Pro and Nano — Gemini has the flexibility to run on everything from data centers to mobile devices. We trained Gemini at scale on our AI-optimized infrastructure using Google's Tensor Processing Units (TPUs) v4 and v5e. Today, we also announced our most powerful and scalable TPU system to date, Cloud TPU v5p.
Gemini is available in some of our core products starting today: Bard is using a fine-tuned version of Gemini Pro for more advanced reasoning, planning, understanding and more. Pixel 8 Pro is the first smartphone engineered for Gemini Nano, using it in features like Summarize in Recorder and Smart Reply in Gboard. And we’re already starting to experiment with Gemini in Search, where it's making our Search Generative Experience (SGE) faster. Early next year, we’ll bring Gemini Ultra to a new Bard Advanced experience. And in the coming months, Gemini will power features in more of our products and services like Ads, Chrome and Duet AI.
Android developers who want to build Gemini-powered apps on-device can now sign up for an early preview of Gemini Nano, our most efficient model, via Android AICore. Starting December 13, developers and enterprise customers will be able to access Gemini Pro via the Gemini API in Vertex AI or Google AI Studio, our free web-based developer tool. And as we continue to refine Gemini Ultra, including completing extensive trust and safety checks, we’ll make it available to select groups before opening it up broadly to developers and enterprise customers early next year.
Explore the collection to learn more about our newest model, and the start of the Gemini era.
ALEX KANTROWITZ, DEC 8, 2023
Google really didn’t have to exaggerate. When the company introduced Gemini this week — its stunning new AI model that beat OpenAI on multiple benchmarks and understands both text and images — it delivered a product that lived up to months of hype. Yet, as the company showed it to the world, it embellished.
In a viral video demonstrating Gemini’s capabilities, Google took significant artistic license. It showed a user speaking with Gemini, but was actually representing a text conversation. It showed a moving video, but was feeding Gemini still images. It showed fast responses, but sped those up. It showed tight model outputs, but shortened them “for brevity.” It showed brief prompts eliciting terrific answers, but listed longer prompts in a blog post.
Exaggerating in tech demos isn’t novel, but Google’s Gemini video felt different. As it circulated among developers and industry watchers this week, the sentiment was near universal: Google, whose research made the generative AI era possible, was pressing.
“It's a little bit shocking that Google, the undisputed pioneer in generative AI, would feel it necessary to juice the results of their demo just to try and one up Microsoft/OpenAI,” said Malcolm Ethridge, an executive vice president at CIC Wealth. “But it also speaks to just how important it is to be seen as a legitimate competitor — let alone be the eventual winner of this arms race in AI.”
Google’s had a weird year. Microsoft used technology it developed — the transformer model — to build a marquee offering with OpenAI. Meta used it to build credibility in the open-source AI movement. NVIDIA used it to add more than $500 billion to its market cap. All these efforts threaten Google’s long-term dominance in search.
Gemini, arriving on the heels of OpenAI’s chaos, was Google’s long-awaited answer. But the pressure to deliver something revolutionary seemed to build, and its marketing lost touch with reality. The unforced error spoke loudly.
Google made the video to “inspire developers,” said Oriol Vinyals, a Deepmind vice president. But it may have had the opposite effect for some. Developers on the tech forum Hacker News kept the video on its front page throughout Thursday, and were not happy about it. “I was fooled,” wrote one user. “It’s one thing to release a hype video with what-ifs and quite another to claim that your new multi-modal model is king of the hill then game all the benchmarks and fake all the demos.”
Vinyals said the prompts and outputs in the video were real, with some shortened. And he posted a video that showed some of the brief prompts working exactly like the longer, more detailed ones in the blog post. But his response didn’t seem to satisfy the masses. “Ah, the age-old art of deceptive demos,” said IBM Fellow Grady Booch, quoting Vinyals. “Dear Google: you should have made this abundantly clear.”
Gemini remains an impressive product. While it won’t roll out fully until next year, it’ll have an advantage serving customers on Google Cloud Platform, many of whom already have their data in the service, as Margins’ Ranjan Roy said in Wednesday’s edition of The Panel. For Google, it’s good to finally have Gemini in the wild.
As for the marketing, while it flopped with developers and the company’s closest watchers, one important constituency seems to have bought in. Alphabet stock jumped 5% following the announcement.
DEC 9, 2023
Just now, from Sam Altman himself
Reality
Let that sink in. ELIZA (built in 1965-1966) beat GPT 3.5. That’s embarrassing! 1966 software that could easily run on my watch running competitively with multi-million GPU clusters trained on a large fraction of the internet. (Full article at https://arxiv.org/abs/2310.20216)
And, sorry Sam, when you actually look at the data. humans are still ahead.
§
But honestly, who cares?
Here’s what I wrote about this almost a decade ago, at The New Yorker, the last time someone tried to hype a result on the Turing Test
and
and very much anticipating the current situation
What I proposed then, as a replacement to the Turing Test, was a Comprehension Challenge; even now, no software would be able to meet that challenge.
Dubious Google Gemini videos aside, nobody is close to passing that yet.
Plus ça change, plus c'est la même chose.
Andrew Tarantola’ Senior Editor
Following a marathon 72-hour debate, European Union legislators Friday have reached a historic deal on its expansive AI Act safety development bill, the broadest-ranging and far-reaching of its kind to date, reports The Washington Post. Details of the deal itself were not immediately available.
"This legislation will represent a standard, a model, for many other jurisdictions out there," Dragoș Tudorache, a Romanian lawmaker co-leading the AI Act negotiation, told The Washington Post, "which means that we have to have an extra duty of care when we draft it because it is going to be an influence for many others."
The proposed regulations would dictate the ways in which future machine learning models could be developed and distributed within the trade bloc, impacting their use in applications ranging from education to employment to healthcare. AI development would be split between four categories depending on how much societal risk each potentially poses — minimal, limited, high, and banned.
Banned uses would include anything that circumvents the user's will, targets protected social groups or provides real-time biometric tracking (like facial recognition). High risk uses include anything "intended to be used as a safety component of a product,” or which are to be used in defined applications like critical infrastructure, education, legal/judicial matters and employee hiring. Chatbots like ChatGPT, Bard and Bing would fall under the "limited risk" metrics.
“The European Commission once again has stepped out in a bold fashion to address emerging technology, just like they had done with data privacy through the GDPR,” Dr. Brandie Nonnecke, Director of the CITRIS Policy Lab at UC Berkeley, told Engadget in 2021. “The proposed regulation is quite interesting in that it is attacking the problem from a risk-based approach,” similar what's been suggested in Canada’s proposed AI regulatory framework.
Ongoing negotiations over the proposed rules had been disrupted in recent weeks by France, Germany and Italy. They were stonewalling talks over the rules guiding how EU member nations could develop Foundational Models, generalized AIs from which more specialized applications can be fine-tuned. OpenAI's GPT-4 is one such foundational model, as ChatGPT, GPTs and other third-party applications are all trained from its base functionality. The trio of countries worried that stringent EU regulations on generative AI models could hamper member nations' efforts to competitively develop them.
The EC had previously addressed the growing challenges of managing emerging AI technologies through an variety of efforts, releasing both the first European Strategy on AI and Coordinated Plan on AI in 2018, followed by the Guidelines for Trustworthy AI in 2019. The following year, the Commission released a White Paper on AI and Report on the safety and liability implications of Artificial Intelligence, the Internet of Things and robotics.
Kyle Wiggers@kyle_l_wiggers / 4:13 PM PST•December 7, 2023
Image Credits: TechCrunch
Grok, a ChatGPT competitor developed by xAI, Elon Musk’s AI startup, has officially launched on X, the site formerly known as Twitter.
Grok began rolling out late this afternoon to X Premium Plus subscribers in the U.S., “Premium Plus” being X’s plan that costs $16 per month for ad-free access to the social network. Longtime subscribers will get priority access to Grok, X said, with the rollout expected to wrap up in the next week.
Grok answers questions conversationally, drawing on a knowledge base similar to that used to train the AI models powering ChatGPT and Google’s Bard. It lives in the X side menu on the web, iOS and Android and can be added to the bottom menu in X’s mobile apps for quicker access.
Grok is underpinned by a generative model called Grok-1, which was trained on data both from the web (up to Q3 2023) and feedback from human assistants. Unlike other chatbots, Grok can also incorporate real-time data from posts on X into its responses, enabling it to answer questions with — in theory — up-to-the-minute info.
The real-time access to X data appears to be a genuine advantage — Grok’s “killer feature,” if you will.
Given a prompt such as “What is happening in AI today?” chatbots like Bard and ChatGPT provide vague, outdated answers that reflect the limits of their training data and filters on their web access. Grok, by contrast, pieces together a response from very recent headlines — although it’s not clear how it’s making its source selections and how often it might hallucinate wrong answers.
Same query on Grok, ChatGPT, and Bard! 🤯
We are vastly underestimating the power of real-time data. Grok nails it!
h/t Scobleizer pic.twitter.com/mWH8FMFxIG
— Bindu Reddy (@bindureddy) December 7, 2023
Asking Grok how many Q Followers there are on X pic.twitter.com/kaHL8GUJ0Z
— MoCheezePlz (@Yeahaboutthat3) December 7, 2023
Apple has open-sourced MLX, a new machine learning framework specifically designed for its Apple silicon chips. Developed by Apple’s machine learning research team, MLX streamlines the process of training and deploying models for researchers working on Mac, iPad, and iPhone.
Just in time for the holidays, we are releasing some new software today from Apple machine learning research.
MLX is an efficient machine learning framework specifically designed for Apple silicon (i.e. your laptop!)
Code: https://t.co/Kbis7IrP80
Docs: https://t.co/CUQb80HGut— Awni Hannun (@awnihannun) December 5, 2023
MLX boasts several features that set it apart from existing frameworks:
Familiar APIs: Python and C++ APIs have familiar frameworks like NumPy and PyTorch, making it easy for experienced researchers to learn.
Effortless Efficiency: MLX uses composable function transformations to optimize Apple silicon performance.
Lazy Computation: Arrays only materialize when needed, preventing unnecessary calculations and boosting resource efficiency.
Dynamic by Design: Computation graphs adapt to input shape changes, simplifying debugging and experimentation.
Multi-Device Powerhouse: MLX seamlessly leverages your Apple device’s CPU and GPU, ensuring you get the most out of your hardware.
Unified Memory Advantage: MLX stores arrays in shared memory, unlike other frameworks, eliminating data movement between devices and further accelerating operations.
Researcher-Friendly: MLX is designed for researchers with a clean and extensible codebase that encourages contributions.
Apple has demonstrated the impressive capabilities of MLX, showcasing its ability to enhance natural language processing through efficient Transformer model training. With LLaMA and LoRA, users can generate large amounts of text on their Apple devices.
Additionally, Stable Diffusion enables stunning image generation. At the same time, OpenAI’s Whisper technology provides accurate and efficient speech recognition directly on Apple devices.
Amanda Silberling @asilbwrites / 10:23 AM PST•December 7, 2023
Image Credits: DBenitostock / Getty Images
It seems like the writing is on the soundproofed wall: The podcast boom is over, and this week’s news is evidence. Spotify laid off 17% of the company — its third round of layoffs this year — and canceled two highly acclaimed shows, including a winner of the Pulitzer Prize for audio reporting. But as a whole, the podcast industry didn’t fail. It’s just that Spotify took a billion-dollar swing and whiffed, and now podcasters have to navigate the fallout.
“Spotify has kind of set the terms of the quote-unquote ‘health’ of the podcasting industry based on their actions as a tech company,” said Eric Silver, co-founder and head of creative at Multitude, an independent podcast collective. “But Spotify’s choices don’t have anything to do with me. It’s just that they keep failing so publicly, and now everyone thinks podcasting is dead, which really frustrates me.”
When outsiders think of podcasting, they may imagine mega-viral hits like “Serial” or long-standing institutions like “This American Life.” But for the long tail of podcast creators — those who make podcasts for a living, but aren’t getting multi-million-dollar deals from Amazon, Apple or Spotify — the industry isn’t as imperiled as it seems. And yet, Spotify’s shadow looms so large over the podcasting industry that it’s impossible for its failures not to reverberate.
In 2021, a year that saw venture capital flowing like champagne at a Gatsby party, Spotify CEO Daniel Ek told Forbes that he wanted his company to be like the Instagram or TikTok of audio.
“Everyone underestimates audio. It should be a multi-hundred-billion-dollar industry,” Ek said at the time. “Audio is ours to win.”
In the last handful of years, we watched as Spotify acquired too many podcasting companies to count — Gimlet, The Ringer, Anchor, Parcast, Megaphone — and then courted big names from Joe Rogan, to Alex Cooper, to Prince Harry with eight- and nine-figure deals. The company dumped over a billion dollars into its efforts to corner podcasting but now has canceled over a dozen shows from the studios it spent so many hundreds of millions to acquire, like Parcast and Gimlet, which have since been combined into one entity and decimated.
“In hindsight, I was too ambitious in investing ahead of our revenue growth,” Ek said after Spotify laid off 600 people in January.
After acquiring Gimlet and Parcast, Spotify made most of the networks’ shows exclusive to the Spotify platform. In theory, this decision would force the listeners of these popular shows to download Spotify in order to keep listening every week — and, hopefully, some of those listeners would convert to paid subscribers. But, according to the Gimlet and Parcast unions, this strategy backfired. Some shows lost more than three-quarters of their audiences after being converted to Spotify exclusives.
“Spotify told show teams that their podcasts were being canceled because of low numbers,” said a joint statement by the Gimlet and Parcast unions, posted after a round of layoffs in October 2022. “But decisions Spotify leadership made directly contributed to these low numbers.”
This isn’t necessarily a bad thing. During the funding boom of 2021, my inbox was flooded with pitches from creator economy startups seeking press about their latest funding rounds. Some of these companies were exciting, but many of them confused me — as a creator myself, I couldn’t imagine myself or my friends in the industry using many of these products. As SignalFire partner Josh Constine told me earlier this year, “Creators aren’t sophisticated enterprise software buyers, nor do they have software integration teams.” In other words, VCs had thrown money at companies that didn’t actually solve any problems for creators’ businesses. So, I wasn’t surprised that when market conditions tightened, the companies that seemed to only want to capitalize on the creator economy hype were no longer being funded.
“A media company needs to have the goal of making enough money for it to survive,” Silver told TechCrunch. This might seem intuitive — surely, a business should try to turn a profit. But that’s not how the world of venture-funded startups works. Spotify, for example, has only reported quarterly profits a handful of times, because its business has prioritized continual growth over returns. The company is not at all unique in that way.
“It’s critical for companies to ‘read the market,’ and right now, the market values efficient growth and doing more with less instead of maximum growth with easy capital,” Creative Juice founder Sima Gandhi told TechCrunch this summer.
This “maximum growth” mindset has poisoned venture-backed digital media companies like Buzzfeed, which descended from a shining star to an IPO embarrassment. The “middle class” of podcasters can’t rely on Spotify, and other media workers can’t rely on failing media conglomerates like G/O Media and Vice anymore. Over the last few years, worker-owned media outlets like Defector, Aftermath and 404 Media have begun cropping up, often founded and staffed by journalists who had been repeatedly laid off from mismanaged media companies. Now the podcasting industry is facing the same reckoning as Spotify’s losses prove that growth can’t take priority over sustainability. Already, podcast studio Maximum Fun has adopted a worker-ownership co-op model, and as podcasters continue to lose trust in big corporations like Spotify, we’ll see this trend continue.
“Spotify is not all of podcasting, although they act as if they are, and make choices as if they’re the only one in the room,” Silver said. “Podcasting is not dead.”
Sarah Perez @sarahpereztc / 8:00 AM PST•December 7, 2023
Image Credits: Guillaume Payen/SOPA Images/LightRocket / Getty Images
In September, Google announced it would shut down its standalone podcasts app, Google Podcasts, sometime next year. Now that the end of 2023 is nearing, the company is today launching a migration tool that will allow U.S. users to shift their existing podcast subscriptions over to YouTube Music, which will be Google’s new home for podcasts.
Users will have plenty of time to export their subscriptions as the official discontinuation of Google Podcasts won’t take place until April 2024. While the current tool only supports U.S. users, Google says it will become available to other markets soon.
The company additionally shared the timeline for the Google Podcasts app’s shutdown. It notes that U.S users will be able to listen to their podcasts in the app through March 2024. And although the app is being discontinued in April, users will still be able to migrate or export subscriptions through July 2024.
The new migration tool will appear in the app in the weeks ahead, through a banner at the top of the screen. Step-by-step instructions will also be documented on Google’s support site. For those who don’t want to move to YouTube Music, an option to export subscriptions to an OPML file will also be provided. This file can be uploaded into any other third-party podcast app that also supports uploads.
Google has been working to make YouTube more of a destination for podcasts for some time, launching a dedicated podcasts homepage last year, and announcing plans this February to bring podcasts to YouTube Music — the company’s Spotify and Apple Music rival.
The change will consolidate Google’s efforts in the audio streaming market by allowing it to combine the listenership that’s today spread across multiple apps. The company had already shuttered Google Play Music years ago, but it took longer to address podcasts.
by Jason Lemkin | Blog Posts
So things aren’t all “bad” in SaaS. Many Cloud leaders stock prices are way, way up in 2023, the Cloud platform leaders have re-accelerated, and leaders like Shopify are having close-to-record years.
Now it sort of has to be that way. So, so many SaaS startups got funded in the Boom, and they just can’t all make it. In fact, Craft’s latest data from the presentation below has seen as many as 80% of Seed startups fail to raise a Series A.
So we’re working through this. But Be Kind. The shutdowns in 2023 have been at a record pace, and it’s only going to continue in 2024. The latest data from Pilot (see below) suggests even more startups will likely run out of runway in 2024. Hopefully, we’ll be through most of it by the end of 2024.
by Jason Lemkin | Blog Posts, Fundraising
SaaS products and services like Pilot track the finances of 1,000s of SaaS and other startup so they’re an interesting source of hard data.
What does Pilot’s latest data say? Something that’s both not surprising but also pretty impactful: 57% of venture-backed startups will have to go “back to market” in 2024 to raise more capital. And 38% have 12 or less months of runway left. Many have already raised a bridge round. And realistically, most won’t have the metrics to pull off another round.
That’s how it works. VCs don’t give startups 10 years of capital. They typically give them just enough to see if it will work, and the startup grows and scales to the next stage. That’s typically been 18 months historically.
At a practical level though, the headlines in 2024 may actually look much worse than 2023 for startup failures. So many were able to cut the burn and stretch their cash through 2023. But you can only cut so much. Ultimately, you also have to grow again to raise more venture capital.
Folks that raised in the Go-Go Times of 2021 in many cases have been able to stretch their cash through 2023. But many will find 2024, those stretches have stretched as far as they can.
SaaS and Cloud growth overall will remain strong. Shopify, Datadog, Crowdstrike, Google Cloud-Azure-AWS, Snowflake, etc. may well put up not just strong numbers, but even stronger than 2023. In fact, Gartner predicts enterprise software spend will cross $1 Trillion Dollars (!) for the first time in 2024!
But with so, so many startups funded in 2021 … we likely will also see a record number shut down in 2024. :(. That’s what this Pilot data also reflects.
So let’s be empathetic to those that wind down in 2024. But also focus on the prize at the same time — record Cloud spend. Carpe Diem.
Will Shanklin, Contributing Reporter’ Fri, Dec 8, 2023
OpenAI’s recent drama hasn’t only caught UK regulators’ attention. Bloomberg reported Friday that the Federal Trade Commission (FTC) is looking into Microsoft’s investment in the Sam Altman-led company and whether it violates US antitrust laws. FTC Chair Lina Khan wrote in a New York Times op-ed earlier this year that “the expanding adoption of AI risks further locking in the market dominance of large incumbent technology firms.”
Bloomberg’s report stresses that the FTC inquiry is preliminary, and the agency hasn’t opened a formal investigation. But Khan and company are reportedly “analyzing the situation and assessing what its options are.” One complicating factor for regulation is that OpenAI is a non-profit, and transactions involving non-corporate entities aren’t required by law to be reported.
In addition, Microsoft’s $13 billion investment doesn’t technically give it control over OpenAI in the eyes of the law, another factor in determining what action a governmental agency might be able to take. However, the recent ousting and re-hiring of Altman — and the integral role Microsoft played in reverting those chess pieces to its preferred positions — suggests the lack of control over the nonprofit is more a technicality than the relationship’s underlying essence.
The UK’s Competition and Markets Authority (CMA) wrote earlier today that it’s considering investigating the relationship between AI’s two dominant players. It said it’s weighing “recent developments,” referring obliquely to the Altman-Microsoft drama. “The CMA will review whether the partnership has resulted in an acquisition of control — that is, where it results in one party having material influence, de facto control or more than 50% of the voting rights over another entity,” the CMA wrote in its news release.
Khan, also challenging Microsoft’s $69 billion Activision Blizzard acquisition, has previously sounded the alarm about the need for AI regulations.
“As these technologies evolve, we are committed to doing our part to uphold America’s longstanding tradition of maintaining the open, fair and competitive markets that have underpinned both breakthrough innovations and our nation’s economic success — without tolerating business models or practices involving the mass exploitation of their users,” the youngest-ever FTC chair wrote in May. “Although these tools are novel, they are not exempt from existing rules, and the F.T.C. will vigorously enforce the laws we are charged with administering, even in this new market.”
Amazon filed a motion on Friday in the Western Washington district court asking a judge to dismiss the Federal Trade Commission’s (FTC) antitrust lawsuit against it. The FTC along with 17 state attorneys general sued Amazon in September, alleging the company uses monopolistic practices that are unfair to both its competitors and consumers. Amazon is now arguing that the FTC did not provide evidence that its practices have driven up prices or harmed consumers, according to Bloomberg.
The FTC’s lawsuit claims Amazon uses illegal tactics to crush its competition — like punishing sellers who list their products for better prices elsewhere by burying them in search results, and coercing sellers into using Amazon’s own fulfillment service by tying it to Prime eligibility. It also accuses Amazon of inflating prices from 2016-2018 using an algorithmic tool codenamed Project Nessie. These increases added up to more than $1 billion, according to the suit.
In Amazon’s motion for dismissal, per AP, Amazon said it’s only engaging in “common retail practices” that “benefit consumers and are the essence of competition.” Amazon attorney Heidi Hubbard wrote that the suit “implausibly, and illogically, assumes that Amazon’s efforts to keep featured prices low on Amazon somehow raised consumer prices across the whole economy,” according to Bloomberg.
Devin Coldewey @techcrunch / 2:06 PM PST•December 4, 2023
Image Credits: Alibaba Group
As if still-image deepfakes aren’t bad enough, we may soon have to contend with generated videos of anyone who dares to put a photo of themselves online: with Animate Anyone, bad actors can puppeteer people better than ever.
The new generative video technique was developed by researchers at Alibaba Group’s Institute for Intelligent Computing. It’s a big step forward from previous image-to-video systems like DisCo and DreamPose, which were impressive all the way back in summer but are now ancient history.
What Animate Anyone can do is not by any means unprecedented, but has passed that difficult space between “janky academic experiment” and “good enough if you don’t look closely.” As we all know, the next stage is just plain “good enough,” where people won’t even bother looking closely because they assume it’s real. That’s where still images and text conversation are currently, wreaking havoc on our sense of reality.
Image-to-video models like this one start by extracting details, like facial feature, patterns and pose, from a reference image like a fashion photo of a model wearing a dress for sale. Then a series of images is created where those details are mapped onto very slightly different poses, which can be motion-captured or themselves extracted from another video.
Previous models showed that this was possible to do, but there were lots of issues. Hallucination was a big problem, as the model has to invent plausible details like how a sleeve or hair might move when a person turns. This leads to a lot of really weird imagery, making the resulting video far from convincing. But the possibility remained, and Animate Anyone is much improved, though still far from perfect.
The technical specifics of the new model are beyond most, but the paper emphasizes a new intermediate step that “enables the model to comprehensively learn the relationship with the reference image in a consistent feature space, which significantly contributes to the improvement of appearance details preservation.” By improving the retention of basic and fine details, generated images down the line have a stronger ground truth to work with and turn out a lot better.
Image Credits: Alibaba Group
They show off their results in a few contexts. Fashion models take on arbitrary poses without deforming or the clothing losing its pattern. A 2D anime figure comes to life and dances convincingly. Lionel Messi makes a few generic movements.
They’re far from perfect — especially about the eyes and hands, which pose particular trouble for generative models. And the poses that are best represented are those closest to the original; if the person turns around, for instance, the model struggles to keep up. But it’s a huge leap over the previous state of the art, which produced way more artifacts or completely lost important details like the color of a person’s hair or their clothing.
It’s unnerving thinking that given a single good-quality image of you, a malicious actor (or producer) could make you do just about anything, and combined with facial animation and voice capture tech, they could also make you express anything at the same time. For now, the tech is too complex and buggy for general use, but things don’t tend to stay that way for long in the AI world.
At least the team isn’t unleashing the code into the world just yet. Though they have a GitHub page, the developers write: “we are actively working on preparing the demo and code for public release. Although we cannot commit to a specific release date at this very moment, please be certain that the intention to provide access to both the demo and our source code is firm.”
Will all hell break loose when the internet is suddenly flooded with dancefakes? We’ll find out, and probably sooner than we would like.
A reminder for new readers. That Was The Week collects the best writing on critical issues in tech, startups, and venture capital. I selected the articles because they are of interest. The selections often include things I entirely disagree with. But they express common opinions, or they provoke me to think. The articles are only snippets. Click on the headline to go to the original. I express my point of view in the editorial and the weekly video below.
This Week’s Video and Podcast:
Content this week from @kteare, @ajkeen, @bennstancil, @kwharrison13, @KateClarkTweets, @GeneTeare, @eringriffith, @ttunguz, @Kantrowitz, @GaryMarcus, @Terrortola, @kyle_l_wiggers, @beriboii, @asilbwrites, @sarahpereztc, @jasonlk, @willshanklin, @_cheymac
Animate Anyone
It’s been a week….
For a start, on my return from travel, I tested positive for Covid 19. Although a test was barely needed, I felt decidedly ill on landing. That was Tuesday; today is Saturday, and the end is in sight. Not the end of the runway, the end of the Covid. I’m out of bed and writing.
But speaking about the end of runways, as depicted in this week’s DALL-E image, we have evidence of the scale of startup closures this week and a very high-profile venture fund closure - Openview from Boston.
As I began writing, The Information published a story about D2IQ - an A16Z-backed unicorn - running out of runway. Openview’s closure was reported a few days earlier.
Kyle Harrison has one of the Essays of the Week, ‘Revisiting the Death of a Venture Fund,’ which views the state of play. He quotes Delian Asparouhov of Founders Fund:
I think venture capitalists are starting to awaken to the idea that ultimately what you need to be aiming for is the most long tail of outcomes. And if you look at the top 5 companies in the NASDAQ right now--Apple, Nvidia, Tesla. People are starting to realize that in order to build these long tail outcome companies you actually need to build things that tend to be a little bit more capex expensive and so that tends to mean that its something that is easier done when you have venture capitalist effectively as a co-founder. And so I think you'll see some more of it, mostly because I think the most interesting companies are going to be more capex intensive over the next decade.
In the context of the end of Runway, he then comments:
Comfort around capital intensity is not the only product that can keep a venture fund alive and thriving. There are a myriad of ways to stand out as a firm. But increasingly, the stakes have been raised. And now its not just a question of facing superiority or irrelevance. It's a function of facing qualification for the right to survive, or death.
Carta covers the accelerating shutdown rate of startups.
And to reinforce the context, we have spoken a lot here about the massive impact that the market reversal of 2021 has had on the value chain of venture capital. 57% of startups on Carta’s platform must raise new capital in 2024. most of them will likely fail for a variety of reasons.
The New York Times piece - From Unicorns to Zombies: Tech Start-Ups Run Out of Time and Money - sums up much of this. Erin Griffith is a wonderful reporter and analyst of the ecosystem. She writes:
Venture investors say that failure is normal and that for every company that goes out of business, there is an outsize success like Facebook or Google. But as many companies that have languished for years now show signs of collapse, investors expect the losses to be more drastic because of how much cash was invested over the last decade.
From 2012 to 2022, investment in private U.S. start-ups ballooned eightfold to $344 billion. The flood of money was driven by low interest rates and successes in social media and mobile apps, propelling venture capital from a cottage financial industry that operated largely on one road in a Silicon Valley town to a formidable global asset class akin to hedge funds or private equity.
During that period, venture capital investing became trendy — even 7-Eleven and “Sesame Street” launched venture funds — and the number of private “unicorn” companies worth $1 billion or more exploded from a few dozen to more than 1,000.
But the advertising profits gushing from the likes of Facebook and Google proved elusive for the next wave of start-ups, which have tried untested business models like gig work, the metaverse, micromobility and cryptocurrencies.
Now some companies are choosing to shut down before they run out of cash, returning what remains to investors. Others are stuck in “zombie” mode — surviving but unable to grow. They can muddle along like that for years, investors said, but will most likely struggle to raise more money.
Jason Lemkin writes that this is normal, and he is right. But we have never had so many companies, with so much sunk capital and so many employees, make up the high percentage of failures as we will in the next period. It should not be a long-term ecosystem concern, but it certainly is a short-term issue.
That said, where it is common for startups to fail, it is rare for venture firms to close down. We will see more of that in 2024 as LPs that invest in funds are reluctant to do so due to a lack of liquidity or over-indexing on venture and needing to rebalance.
Openview is a staggering closure as it still has over $1bn of assets under management. Some partners do not want to spend the next several years managing an over-valued portfolio on the last round through the inevitable corrected valuations and closures that will produce.
By contrast, Calpers - the California Public Employees Retirement Fund - has significantly increased its commitments to venture capital. As always, more than one trend is in play simultaneously. Out of death comes growth and renewal.
Talking of that, OpenAI’s over-reported death and rebirth has spawned a new aggressive marketing campaign from Google for its Gemini competitor. Perhaps too aggressive. A characteristically understated Google stepped over lines in producing a highly edited video showing capabilities Gemini does not possess - at least not as depicted — an extensive and free win for OpenAI.
That said, Google Gemini is a decent product. Like others, it is highly constrained due to Google’s paranoia about doing things that would draw unwanted attention. Its web reading is somewhat limited. It seems able to be given a single URL to summarize. However, it cannot read RSS feeds or crawl links. It has been constrained when it comes to politics. Imagine if we limited our children’s brains in this way…. Oh, wait.. :-).
And more regulation news, which seems to grow each week. The EU agrees on an AI regulation act destined to render the continent irrelevant as builders engineer the next version of human ingenuity. The FTC wants to look into Microsoft’s OpenAI investment and will have as much luck as it has had with its many 2023 failed efforts with other companies.
Apple releasing its AI tools as open source got some attention but will likely do more than the FTC in curbing AI centralization.
My editorials are not surveys. They are opinions, so it behooves me to state that I believe the new green shoots in 2023 are more significant signposts to the future than the gloomy stories in this week’s edition. Google will produce a compelling AI platform, and OpenAI will go on to even greater things. A blossoming of new applications leveraging AI developments will revolutionize work and the human learning and building experience. Optimism is the proper framework for understanding current trends.
OK, enough said. My Covid recovery wants me to stop writing so I am out of Runway, too — more in this week’s video.
DEC 8, 2023
Six months ago, the modern data stack hit an important milestone: It was big enough for Microsoft to care about it. Until this year, Microsoft had operated in a different cinematic universe than the classic modern data stack vendors1—in enterprises versus startups, in Houston and Chicago versus Silicon Valley, and in companies that choose software based on its features and cost versus those that choose Notion based on its aesthetic and vibe.2 This May, however, Microsoft stopped ignoring the modern data stack, and decided to fight it:
Microsoft today launched Microsoft Fabric, a new end-to-end data and analytics platform. ... “Over the last five to 10 years, there has been a pretty massive level of innovation—which is great and that’s awesome because there’s lots of new technologies out there—but it’s also caused a lot of fragmentation of the modern data stack,” Arun Ulag, Microsoft’s corporate VP for Azure Data, told me. ... “When I talk to customers, one of the messages I hear consistently is that they’re tired of paying this integration tax,” he said.
So Microsoft looked at the core data analytics workloads (data integration, engineering, warehousing, data science, real-time analytics and business intelligence) and looked at how it could build a unified experience around this. … [With Fabric, there’s] “literally just one thing to buy, and it allows customers to save a lot of costs, which, especially in today’s environment, is really important,” said Ulag.
For the modern data stack’s ascendent clique, this was alway the risk. Today’s data tools are famously warehouse-centric; what happens if the warehouse providers decide that they want to build their own versions? Databases make a lot more money than the applications that run on top of them; providers of the latter sell at the pleasure of the former. What if Snowflake gives a native reporting tool away to all its customers? Or Databricks builds (or buys) dbt, and provides it for no extra charge? Or Microsoft bundles a cheaper Fivetran on top of their Azure SQL databases? Despite what the inspirational posters say, when Microsoft stops ignoring you and starts fighting you, they usually win.3
When it first came out, I thought this was what Fabric was about—the modern data stack consolidating around the warehouse at its center.4 The cash cows were taking control of their loss leaders. As I said at the time, even if Microsoft didn’t try to directly poach every Astronomer or Hightouch customer, “the modern data stack may not get steamrolled, but stunted, frozen out of the top of the market by a Microsoft sales team that already has a standing tee time with every enterprise CIO."
That, it turns out, may have been delusional main-character energy. For better or for worse, Microsoft’s fight’s not with us,5 but for something much bigger.
If Redshift were to write a memoir, it would be a weird book. When Amazon launched Redshift in 2012, it was the first affordable cloud-based data warehouse, and an instant success. Practically overnight, an entire ecosystem of data companies formed around it. Segment’s first destination for writing events was Redshift; Fivetran’s first users wanted to combine all of their analytics data in Redshift; people ran their businesses out of Redshift.
And then, at the height of its fame, Amazon sold it out. In the late 2010s, AWS’ sales reps started getting paid for selling Snowflake too, so long as it was hosted on AWS. Snowflake grew, Redshift faded, and the modern data stack’s first star became a has-been with an interesting backstory and no future.
A couple years ago, Erik Bernhardsson wondered why Redshift’s parents traded it away. His answer, backed up by research and math and the sort of concrete analysis that this blog prefers not to bother with, was that Amazon probably makes more money—and definitely makes easier money—by selling infrastructure to one big vendor like Snowflake than it does by selling databases to thousands of individual buyers like it had to with Redshift. Customers pay Snowflake, Snowflake pays AWS, and “AWS basically ends up with the same bottom line impact, but effectively ‘outsources’ to Snowflake all the cost of building software and selling it. That seems like a good deal for them!”
In the hierarchy of things you want to sell, databases aren’t actually the cash cow, but the middle fish—bigger than the apps, but smaller than the cloud. Just as BI tools and ETL companies sell at the pleasure of database vendors, database vendors sell at the pleasure of the cloud providers. Databricks can undercut the data catalog market by offering one for free to its customers—but Google can undercut Databricks by offering big BigQuery6 discounts to customers who use GCP. Frank Slootman comes for data companies; Satya Nadella comes for everyone. If database vendors want to build durable defenses against their larger predators, eating the modern data stack doesn’t help all that much. They need to prioritize a higher order bit.7
Easier said than done, though. If companies could just decide to build an $80-billion-a-year cloud infrastructure business, AWS wouldn’t be an $80-billion-a-year cloud infrastructure business. But what if you could build a different higher order bit, in a new market that doesn’t yet have any dominant incumbents? As I said nine months ago, there might be one out there for the taking:
In ten years, a new type of cloud—a generative one, a commercial Skynet, a public imagination—will undergird nearly every piece of technology we use. … Just as cloud providers built out hundreds of utilities that are all underpinned by core services like EC2, I'd expect OpenAI to do the same thing on top of GPT and other foundational models. …
Public AI providers like OpenAI would become another backbone for the internet. Nearly every piece of technology will rely on their models. Nearly every piece of technology will rely on their models. Outlook will need them to summarize our emails. Github will use them to automate code reviews. DoorDash will need them to help guide you through your order. Delta will depend on them for booking flights. Facebook might not be able to open doors without them. …
The final equilibrium is the same as it for the cloud providers: A few companies win the market, and the rest of us come to accept their bills as the cost of doing business.
Assuming this whole AI thing isn’t overblown,8 I think this still holds up, with one major adjustment. In that post, I said the public AI providers would sell inference: Developers would send OpenAI a prompt or some sort of media, and OpenAI would send back a response. “Training and running LLMs is very expensive,” I thought, “like building data centers is expensive.”
The part seems partially wrong. It is expensive, but it’s not that expensive. Moreover, lots of large companies want to train their own models, both to get better results and to maintain control of their data. And the market for that could be huge: Companies are already spending more than a billion dollars a year with OpenAI—despite it being a new technology that most people are just experimenting with, OpenAI serving generic models, and companies having to deal with the security problems associated with sharing data with a third party.
KYLE HARRISON, DEC 9, 2023
I know a lot of people who were obsessed with Game of Thrones. And I know a lot of people who were disappointed by the ending. I never watched the show, but I know that one of the things a lot of fans came to expect was a slew of surprising deaths.
While I can't empathize with the laundry list of medieval surprise deaths, I can relate with the experience having watched The Walking Dead for almost 10 seasons. The show would repeatedly, and gratuitously, kill off characters. Often, characters that you thought they could never do without.
That feeling of week-after-week surprises, many of them gut punches, is how I feel about the news cycle in the world of tech. From the collapse of SVB to the firing of Sam Altman to the return of Sam Altman, it's been one hit after the other. This week was no exception.
For those of you who didn't see the news, OpenView Venture Partners based in Boston abruptly announced that it was laying off the majority of its staff and shutting down new investments. This isn't a first time fund failing to raise a second, or a fringey crowdfunding idea that failed to take shape.
This is a nearly-twenty year old firm with $2.4B of AUM that has invested in companies like Calendly, Datadog, and Expensify. This is a firm that has had some of its investors listed on the Midas List. Of all the venture firms you would expect to shutdown, this was certainly not on my bingo card. So what happened?
As far as all the reporting I've read, this was a case of a partnership that broke down. The firm had previously been led by its founder, Scott Maxwell, and three other general partners — Mackey Craven, Ricky Pelletier, and Blake Bartlett. Apparently quite suddenly, both Pelletier and Craven decided to leave the firm.
Did the firms most recent fund have crappy-looking returns? Sure. It was mostly deployed in 2021. Did the firm have a key man clause that could trigger LP action if certain folks left? Sure. Most firms do. But were the returns uniquely bad? Definitely not. No worse than some jungle cat-themed crossover folks we know. Was the key man clause triggered? No. Maxwell and Bartlett are still with the firm.
So how does a venture fund die? Lucky for me, I've already written about this in detail.
In May 2022, I wrote a piece called "The Death of a Venture Fund" in which I unpacked the ways a VC firm can die, albeit a typically slow death, certainly compared to OpenView's shutdown seemingly within a few days. Some of the biggest drivers? Performance, playing it safe, succession planning, bad timing, or reputation destruction. Almost all of those things take time to kill a venture fund. Outside of outright fraud, its very rare for a venture fund to throw in the towel.
One of the reasons that venture funds are so gosh darn hard to kill is because they have very long lifecycles. A typical venture fund has a lifecycle of 10 years, usually 1-3 years to deploy, and 7-9 years to harvest. As a result, even if you shut down today, you'll still have positions to manage whether you like it or not. For example, OpenView still has ~$1B of deployed capital that they have to manage whether they want to die or not.
There are some firms who revolve their whole business around buying out the portfolios of distress venture funds. Tiger tried to do this with Evercore selling all or some of its positions in an effort to generate near-term liquidity. The biggest obstacle is a main staple of venture capital as an asset class — it's all illiquid. That makes it a much harder asset to move than stocks or bonds.
While this specific firm shutdown may have caught a lot of folks off guard (not the least of which was probably many of the people on the OpenView team), the marked increase in VC shutdowns wasn't a surprise to everyone. Two of the best predictions I saw last year:
The first came from Josh Wolfe at Lux:
And the second came from Frank Rottman at QED:
So what were some of the forces that these investors were seeing? Frank's presentation on the three-body problem in venture was particularly exhaustive. He points to several changing forces that have dramatically transformed the venture landscape:
Massive explosion in the amount of funding
Massive increases in value creation (M&A, IPOs, etc.) (I wrote about this in "What's In A Valuation?")
Increased exposure to playbooks for building startups (I wrote about this in "The Professionalization of Startups")
Increased allocation to private market capital deployment
But as the broader macroeconomic picture has shifted dramatically, each of these forces have been significantly impacted. The amount of funding receded, the value of the outcomes shrank, and the allocation amounts directed to private markets were called into question.
But what didn't change was the knowledge and expertise around how to build companies. Granted, a whole chunk of pages in that playbook became invalidated (I'm looking at you blitzscaling). But the expertise remains. Now the question is what does the market for financial support beams look like that sits adjacent to this playbook?
More recently, Alexander Niehenke wrote a great thread unpacking what is happening in venture.
He makes a similar point to Rotman in how things have now changed:
"The venture business model is "share the pizza pie" which works great when the pie is growing as it has over the past decade. Suddenly that pizza pie is shrinking (smaller funds, longer fundraising cycles, less returns) meaning there are less fees & carry going around."
So what do you do? In Rotman's presentation, he talks about a retreat to "distinct points of stability." In his estimation, those points of stability are:
Scale
Non-consensus alpha
Late-stage generalist
Solo capitalist
I would encourage you to go read his whole presentation because it's quite good, and very well thought out. I won't touch on every point, but one point that he makes is one that I've made several times. This idea of the value of scale. Scale brings with it an increased likelihood of seeing deal flow, the ability to balance speed with thoroughness, and an enriched support function that can be subsidized by the fees available at scale.
By Kate Clark Dec. 7, 2023 2:52 PM PST
Here’s a number that might surprise you: In the first six months of the year, California Public Employees’ Retirement System committed about $4.5 billion to venture capital funds, according to public filings. That’s almost 15% of the total capital raised by U.S. VC firms in the same period, and a staggering increase over Calpers’ 2022 commitments to VC funds, which totaled about $1.5 billion.
As most LPs take a step back from risky VC fund investments, California’s biggest pension fund is throwing cash at Sand Hill Road heavyweights. Why? Because it’s fearful of repeating past mistakes. In the wake of the 2008 financial crisis, a period Calpers dubbed “the lost decade,” the pension fund retreated from private equity, a strategy shift that lost it $11 billion in returns, it estimates.
This year so far, the pension fund has committed $300 million to Lux Capital’s eighth fund, $500 million to Thrive Capital’s eighth fund, $400 million to a new Andressen Horowitz fund and $700 million across several funds associated with General Catalyst, according to the filings. Calpers hasn’t yet disclosed the investments it made during the second half of the year.
Many VC firms have avoided raising from public pension funds so they can keep their returns private. But with the number of VC funds raised this year on pace to hit a nine-year low, according to PitchBook, firms have opened themselves up to almost anyone willing to write a check. Calpers’ investment in Andreessen Horowitz, first reported by Fortune, represents the pension fund’s first investment in the tech fund. It also made its first investment in Thrive Capital earlier this year, as I first reported.
As one LP in VC firms put it to me, in this market, “all money is green.”
The $4.5 billion figure likely makes Calpers one of the biggest backers of U.S. VC firms, and that doesn’t even include the fund’s total PE commitments. Calpers has put billions in larger PE funds belonging to TPG, Silver Lake and others. In total, it committed around $10 billion to PE in the first two quarters, a massive increase from its historical norm. From 2009 to 2018, Calpers put $5 billion or less in new money into PE (including VC) each year, The Wall Street Journal reported.
Calpers’ 2022 VC commitments include HongShan (formerly known as Sequoia Capital China), Lightspeed Venture Partners and Bessemer Venture Partners. This year, it also made commitments to Valor Equity Partners, Insight Partners, Bain Capital Ventures, Canaan Partners and Mayfield Fund.
The pension fund is undergoing big changes. In September, its chief investment officer, Nicole Musicco, stepped down after only 18 months. Meanwhile, it’s employing a bold new private investments strategy led by Anton Orlich, who joined Calpers just last year from health insurer Kaiser Permanente, where he was head of alternative investments.
Calpers is expected to approve an increase of its PE allocation to 17% in March, according to public documents. “It is very possible that in two years’ time we would look to raise that again,” Calpers interim Chief Investment Officer Dan Bienvenue said in a recent Calpers board meeting.
This should all be welcome news for venture capitalists. Total VC fundraising dropped 72% in the first three quarters, compared to the same period last year. And raising money is expected to be even more challenging next year.
But other investors in VC funds are privately sniping about Calpers’ aggressive approach. LPs “with real discipline are not making a lot of new VC fund investments,” another LP said.
Gené Teare, December 5, 2023, @geneteare
Global venture funding reached $19.2 billion in November 2023, down marginally month over month, Crunchbase data shows. Funding fell around 16% from the $23 billion invested in November 2022, which was already down by two-thirds from November 2021.
Early-stage funding declined the most year over year — falling 34% — an indication that venture investors continue to scale back even when investing in younger startups. Seed funding slowed more than 15%.
In contrast, late-stage funding increased by around 7% compared to November 2022.
All told, venture funding to U.S. companies in November reached close to $9 billion — slightly less than 50% of total global funding.
Health care and financial services companies raised the largest amounts last month, with more than $3 billion invested in each of those sectors.
Artificial intelligence companies raised $2.4 billion in total in November. The biggest fundings went to companies in the AI infrastructure sector: Germany-based Aleph Alpha and Silicon Valley-based Together AI.
The creation and destruction of value were on full display last month.
The trial of Sam Bankman-Fried delivered a guilty verdict on Nov. 2, a year after the collapse of FTX, the cryptocurrency exchange he founded, which at its peak in early 2022 was valued at $32 billion.
OpenAI also faced what looked like an existential threat with the firing of Sam Altman and more than 90% of employees threatening to quit. That threat ultimately didn’t come to pass, as Altman was rehired within a week. OpenAI was most recently valued at $29 billion earlier this year, with a secondary offering in the works for employees that is expected to place the value of the company at around $80 billion.
Other large valuations in November included Blockchain.com raising funding at a $7 billion valuation, half of its previous valuation. Israeli-based Next Insurance raised strategic capital at a flat valuation from its 2021 funding at around $4 billion — viewed as an indication of strength in this funding environment.
Gené Teare, December 8, 2023, @geneteare
Only nine companies joined The Crunchbase Unicorn Board in November, together adding $11.7 billion in value to the board, amid a slower funding environment for startups.
Financial services dominated the list of new unicorns minted last month, with three companies — in buy now, pay later; rebate management; and lending — from the sector galloping onto the board.
The companies were from across the globe. Two new unicorns each joined from the U.S. and China. The U.K., India, Saudi Arabia and the Cayman Islands each contributed one.
So far in 2023, 86 companies have joined Crunchbase’s unicorn list. This compares to 318 companies that joined in 2022 and 617 during the peak funding market in 2021.
Here are the companies that joined the Unicorn Board in November, by sector.
Buy now, pay later shopping app Tabby, based in Saudi Arabia, raised a $200 million Series D funding at a $1.5 billion valuation. The shopping app operates across the MENA market. The funding was led by Wellington Management.
Business rebate management tool Enable, which was founded in London and is now headquartered in San Francisco, raised a $120 million Series D funding at a value of $1.1 billion. The round was led by Lightspeed Venture Partners.
India-based lending platform InCred raised a $60 million Series D funding at a billion-dollar valuation. The company provides loans to consumers, small and medium-sized businesses and students.
Cayman Islands-based Wormhole, a blockchain messaging tool, raised $225 million in the form of token warrants at a $2.5 billion valuation, the largest round in the Web3 sector this year.
Wuhan, China-based Xingji Meizu is a developer of smartphone technology to empower smart cars. It raised $276 million at a $1.4 billion valuation. The funding was led by Harvest Global Investments and Asia Investment Fund.
U.K.-based Castore, a tech-enabled sportswear brand, raised a $184 million private equity round led by Raine Ventures. The company was valued at $1.2 billion in the deal.
Israel-based Cybersecurity BioCatch, which helps banks and fintech identify fraud and money laundering from biometric data, raised a $70 million secondary market transaction led by Sapphire Ventures that valued the company at $1 billion.
Beijing-based 01.AI, an open-source large-language AI company, raised an undisclosed funding amount at a value of $1 billion. The funding was led by Alibaba Cloud. Notably, 01.AI reached this value within eight months of its founding.
Chinese smart-device developer Rokid, headquartered in Redwood City, California, raised a strategic funding from NetDragon which valued the company at $1 billion.
After staving off collapse by cutting costs, many young tech companies are out of options, fueling a cash bonfire.
Reporting from San Francisco
Dec. 7, 2023
WeWork raised more than $11 billion in funding as a private company. Olive AI, a health care start-up, gathered $852 million. Convoy, a freight start-up, raised $900 million. And Veev, a home construction start-up, amassed $647 million.
In the last six weeks, they all filed for bankruptcy or shut down. They are the most recent failures in a tech start-up collapse that investors say is only beginning.
After staving off mass failure by cutting costs over the past two years, many once-promising tech companies are now on the verge of running out of time and money. They face a harsh reality: Investors are no longer interested in promises. Rather, venture capital firms are deciding which young companies are worth saving and urging others to shut down or sell.
It has fueled an astonishing cash bonfire. In August, Hopin, a start-up that raised more than $1.6 billion and was once valued at $7.6 billion, sold its main business for just $15 million. Last month, Zeus Living, a real estate start-up that raised $150 million, said it was shutting down. Plastiq, a financial technology start-up that raised $226 million, went bankrupt in May. In September, Bird, a scooter company that raised $776 million, was delisted from the New York Stock Exchange because of its low stock price. Its $7 million market capitalization is less than the value of the $22 million Miami mansion that its founder, Travis VanderZ anden, bought in 2021.
“As an industry we should all be braced to hear about a lot more failures,” said Jenny Lefcourt, an investor at Freestyle Capital. “The more money people got before the party ended, the longer the hangover.”
Getting a full picture of the losses is difficult since private tech companies are not required to disclose when they go out of business or sell. The industry’s gloom has also been masked by a boom in companies focused on artificial intelligence, which has attracted hype and funding over the last year.
But approximately 3,200 private venture-backed U.S. companies have gone out of business this year, according to data compiled for The New York Times by PitchBook, which tracks start-ups. Those companies had raised $27.2 billion in venture funding. PitchBook said the data was not comprehensive and probably undercounts the total because many companies go out of business quietly. It also excluded many of the largest failures that went public, such as WeWork, or that found buyers, like Hopin.
Carta, a company that provides financial services for many Silicon Valley start-ups, said 87 of the start-ups on its platform that raised at least $10 million had shut down this year as of October, twice the number for all of 2022.
This year has been “the most difficult year for start-ups in at least a decade,” Peter Walker, Carta’s head of insights, wrote on LinkedIn.
Venture investors say that failure is normal and that for every company that goes out of business, there is an outsize success like Facebook or Google. But as many companies that have languished for years now show signs of collapse, investors expect the losses to be more drastic because of how much cash was invested over the last decade.
From 2012 to 2022, investment in private U.S. start-ups ballooned eightfold to $344 billion. The flood of money was driven by low interest rates and successes in social media and mobile apps, propelling venture capital from a cottage financial industry that operated largely on one road in a Silicon Valley town to a formidable global asset class akin to hedge funds or private equity.
During that period, venture capital investing became trendy — even 7-Eleven and “Sesame Street” launched venture funds — and the number of private “unicorn” companies worth $1 billion or more exploded from a few dozen to more than 1,000.
But the advertising profits gushing from the likes of Facebook and Google proved elusive for the next wave of start-ups, which have tried untested business models like gig work, the metaverse, micromobility and cryptocurrencies.
Now some companies are choosing to shut down before they run out of cash, returning what remains to investors. Others are stuck in “zombie” mode — surviving but unable to grow. They can muddle along like that for years, investors said, but will most likely struggle to raise more money.
Dec 06, 2023
Today we introduced Gemini, our largest and most capable AI model — and the next step on our journey toward making AI helpful for everyone. Built from the ground up to be multimodal, Gemini can generalize and seamlessly understand, operate across and combine different types of information, including text, images, audio, video and code. This means it has sophisticated multimodal reasoning and advanced coding capabilities. And with three different sizes — Ultra, Pro and Nano — Gemini has the flexibility to run on everything from data centers to mobile devices. We trained Gemini at scale on our AI-optimized infrastructure using Google's Tensor Processing Units (TPUs) v4 and v5e. Today, we also announced our most powerful and scalable TPU system to date, Cloud TPU v5p.
Gemini is available in some of our core products starting today: Bard is using a fine-tuned version of Gemini Pro for more advanced reasoning, planning, understanding and more. Pixel 8 Pro is the first smartphone engineered for Gemini Nano, using it in features like Summarize in Recorder and Smart Reply in Gboard. And we’re already starting to experiment with Gemini in Search, where it's making our Search Generative Experience (SGE) faster. Early next year, we’ll bring Gemini Ultra to a new Bard Advanced experience. And in the coming months, Gemini will power features in more of our products and services like Ads, Chrome and Duet AI.
Android developers who want to build Gemini-powered apps on-device can now sign up for an early preview of Gemini Nano, our most efficient model, via Android AICore. Starting December 13, developers and enterprise customers will be able to access Gemini Pro via the Gemini API in Vertex AI or Google AI Studio, our free web-based developer tool. And as we continue to refine Gemini Ultra, including completing extensive trust and safety checks, we’ll make it available to select groups before opening it up broadly to developers and enterprise customers early next year.
Explore the collection to learn more about our newest model, and the start of the Gemini era.
ALEX KANTROWITZ, DEC 8, 2023
Google really didn’t have to exaggerate. When the company introduced Gemini this week — its stunning new AI model that beat OpenAI on multiple benchmarks and understands both text and images — it delivered a product that lived up to months of hype. Yet, as the company showed it to the world, it embellished.
In a viral video demonstrating Gemini’s capabilities, Google took significant artistic license. It showed a user speaking with Gemini, but was actually representing a text conversation. It showed a moving video, but was feeding Gemini still images. It showed fast responses, but sped those up. It showed tight model outputs, but shortened them “for brevity.” It showed brief prompts eliciting terrific answers, but listed longer prompts in a blog post.
Exaggerating in tech demos isn’t novel, but Google’s Gemini video felt different. As it circulated among developers and industry watchers this week, the sentiment was near universal: Google, whose research made the generative AI era possible, was pressing.
“It's a little bit shocking that Google, the undisputed pioneer in generative AI, would feel it necessary to juice the results of their demo just to try and one up Microsoft/OpenAI,” said Malcolm Ethridge, an executive vice president at CIC Wealth. “But it also speaks to just how important it is to be seen as a legitimate competitor — let alone be the eventual winner of this arms race in AI.”
Google’s had a weird year. Microsoft used technology it developed — the transformer model — to build a marquee offering with OpenAI. Meta used it to build credibility in the open-source AI movement. NVIDIA used it to add more than $500 billion to its market cap. All these efforts threaten Google’s long-term dominance in search.
Gemini, arriving on the heels of OpenAI’s chaos, was Google’s long-awaited answer. But the pressure to deliver something revolutionary seemed to build, and its marketing lost touch with reality. The unforced error spoke loudly.
Google made the video to “inspire developers,” said Oriol Vinyals, a Deepmind vice president. But it may have had the opposite effect for some. Developers on the tech forum Hacker News kept the video on its front page throughout Thursday, and were not happy about it. “I was fooled,” wrote one user. “It’s one thing to release a hype video with what-ifs and quite another to claim that your new multi-modal model is king of the hill then game all the benchmarks and fake all the demos.”
Vinyals said the prompts and outputs in the video were real, with some shortened. And he posted a video that showed some of the brief prompts working exactly like the longer, more detailed ones in the blog post. But his response didn’t seem to satisfy the masses. “Ah, the age-old art of deceptive demos,” said IBM Fellow Grady Booch, quoting Vinyals. “Dear Google: you should have made this abundantly clear.”
Gemini remains an impressive product. While it won’t roll out fully until next year, it’ll have an advantage serving customers on Google Cloud Platform, many of whom already have their data in the service, as Margins’ Ranjan Roy said in Wednesday’s edition of The Panel. For Google, it’s good to finally have Gemini in the wild.
As for the marketing, while it flopped with developers and the company’s closest watchers, one important constituency seems to have bought in. Alphabet stock jumped 5% following the announcement.
DEC 9, 2023
Just now, from Sam Altman himself
Reality
Let that sink in. ELIZA (built in 1965-1966) beat GPT 3.5. That’s embarrassing! 1966 software that could easily run on my watch running competitively with multi-million GPU clusters trained on a large fraction of the internet. (Full article at https://arxiv.org/abs/2310.20216)
And, sorry Sam, when you actually look at the data. humans are still ahead.
§
But honestly, who cares?
Here’s what I wrote about this almost a decade ago, at The New Yorker, the last time someone tried to hype a result on the Turing Test
and
and very much anticipating the current situation
What I proposed then, as a replacement to the Turing Test, was a Comprehension Challenge; even now, no software would be able to meet that challenge.
Dubious Google Gemini videos aside, nobody is close to passing that yet.
Plus ça change, plus c'est la même chose.
Andrew Tarantola’ Senior Editor
Following a marathon 72-hour debate, European Union legislators Friday have reached a historic deal on its expansive AI Act safety development bill, the broadest-ranging and far-reaching of its kind to date, reports The Washington Post. Details of the deal itself were not immediately available.
"This legislation will represent a standard, a model, for many other jurisdictions out there," Dragoș Tudorache, a Romanian lawmaker co-leading the AI Act negotiation, told The Washington Post, "which means that we have to have an extra duty of care when we draft it because it is going to be an influence for many others."
The proposed regulations would dictate the ways in which future machine learning models could be developed and distributed within the trade bloc, impacting their use in applications ranging from education to employment to healthcare. AI development would be split between four categories depending on how much societal risk each potentially poses — minimal, limited, high, and banned.
Banned uses would include anything that circumvents the user's will, targets protected social groups or provides real-time biometric tracking (like facial recognition). High risk uses include anything "intended to be used as a safety component of a product,” or which are to be used in defined applications like critical infrastructure, education, legal/judicial matters and employee hiring. Chatbots like ChatGPT, Bard and Bing would fall under the "limited risk" metrics.
“The European Commission once again has stepped out in a bold fashion to address emerging technology, just like they had done with data privacy through the GDPR,” Dr. Brandie Nonnecke, Director of the CITRIS Policy Lab at UC Berkeley, told Engadget in 2021. “The proposed regulation is quite interesting in that it is attacking the problem from a risk-based approach,” similar what's been suggested in Canada’s proposed AI regulatory framework.
Ongoing negotiations over the proposed rules had been disrupted in recent weeks by France, Germany and Italy. They were stonewalling talks over the rules guiding how EU member nations could develop Foundational Models, generalized AIs from which more specialized applications can be fine-tuned. OpenAI's GPT-4 is one such foundational model, as ChatGPT, GPTs and other third-party applications are all trained from its base functionality. The trio of countries worried that stringent EU regulations on generative AI models could hamper member nations' efforts to competitively develop them.
The EC had previously addressed the growing challenges of managing emerging AI technologies through an variety of efforts, releasing both the first European Strategy on AI and Coordinated Plan on AI in 2018, followed by the Guidelines for Trustworthy AI in 2019. The following year, the Commission released a White Paper on AI and Report on the safety and liability implications of Artificial Intelligence, the Internet of Things and robotics.
Kyle Wiggers@kyle_l_wiggers / 4:13 PM PST•December 7, 2023
Image Credits: TechCrunch
Grok, a ChatGPT competitor developed by xAI, Elon Musk’s AI startup, has officially launched on X, the site formerly known as Twitter.
Grok began rolling out late this afternoon to X Premium Plus subscribers in the U.S., “Premium Plus” being X’s plan that costs $16 per month for ad-free access to the social network. Longtime subscribers will get priority access to Grok, X said, with the rollout expected to wrap up in the next week.
Grok answers questions conversationally, drawing on a knowledge base similar to that used to train the AI models powering ChatGPT and Google’s Bard. It lives in the X side menu on the web, iOS and Android and can be added to the bottom menu in X’s mobile apps for quicker access.
Grok is underpinned by a generative model called Grok-1, which was trained on data both from the web (up to Q3 2023) and feedback from human assistants. Unlike other chatbots, Grok can also incorporate real-time data from posts on X into its responses, enabling it to answer questions with — in theory — up-to-the-minute info.
The real-time access to X data appears to be a genuine advantage — Grok’s “killer feature,” if you will.
Given a prompt such as “What is happening in AI today?” chatbots like Bard and ChatGPT provide vague, outdated answers that reflect the limits of their training data and filters on their web access. Grok, by contrast, pieces together a response from very recent headlines — although it’s not clear how it’s making its source selections and how often it might hallucinate wrong answers.
Same query on Grok, ChatGPT, and Bard! 🤯
We are vastly underestimating the power of real-time data. Grok nails it!
h/t Scobleizer pic.twitter.com/mWH8FMFxIG
— Bindu Reddy (@bindureddy) December 7, 2023
Asking Grok how many Q Followers there are on X pic.twitter.com/kaHL8GUJ0Z
— MoCheezePlz (@Yeahaboutthat3) December 7, 2023
Apple has open-sourced MLX, a new machine learning framework specifically designed for its Apple silicon chips. Developed by Apple’s machine learning research team, MLX streamlines the process of training and deploying models for researchers working on Mac, iPad, and iPhone.
Just in time for the holidays, we are releasing some new software today from Apple machine learning research.
MLX is an efficient machine learning framework specifically designed for Apple silicon (i.e. your laptop!)
Code: https://t.co/Kbis7IrP80
Docs: https://t.co/CUQb80HGut— Awni Hannun (@awnihannun) December 5, 2023
MLX boasts several features that set it apart from existing frameworks:
Familiar APIs: Python and C++ APIs have familiar frameworks like NumPy and PyTorch, making it easy for experienced researchers to learn.
Effortless Efficiency: MLX uses composable function transformations to optimize Apple silicon performance.
Lazy Computation: Arrays only materialize when needed, preventing unnecessary calculations and boosting resource efficiency.
Dynamic by Design: Computation graphs adapt to input shape changes, simplifying debugging and experimentation.
Multi-Device Powerhouse: MLX seamlessly leverages your Apple device’s CPU and GPU, ensuring you get the most out of your hardware.
Unified Memory Advantage: MLX stores arrays in shared memory, unlike other frameworks, eliminating data movement between devices and further accelerating operations.
Researcher-Friendly: MLX is designed for researchers with a clean and extensible codebase that encourages contributions.
Apple has demonstrated the impressive capabilities of MLX, showcasing its ability to enhance natural language processing through efficient Transformer model training. With LLaMA and LoRA, users can generate large amounts of text on their Apple devices.
Additionally, Stable Diffusion enables stunning image generation. At the same time, OpenAI’s Whisper technology provides accurate and efficient speech recognition directly on Apple devices.
Amanda Silberling @asilbwrites / 10:23 AM PST•December 7, 2023
Image Credits: DBenitostock / Getty Images
It seems like the writing is on the soundproofed wall: The podcast boom is over, and this week’s news is evidence. Spotify laid off 17% of the company — its third round of layoffs this year — and canceled two highly acclaimed shows, including a winner of the Pulitzer Prize for audio reporting. But as a whole, the podcast industry didn’t fail. It’s just that Spotify took a billion-dollar swing and whiffed, and now podcasters have to navigate the fallout.
“Spotify has kind of set the terms of the quote-unquote ‘health’ of the podcasting industry based on their actions as a tech company,” said Eric Silver, co-founder and head of creative at Multitude, an independent podcast collective. “But Spotify’s choices don’t have anything to do with me. It’s just that they keep failing so publicly, and now everyone thinks podcasting is dead, which really frustrates me.”
When outsiders think of podcasting, they may imagine mega-viral hits like “Serial” or long-standing institutions like “This American Life.” But for the long tail of podcast creators — those who make podcasts for a living, but aren’t getting multi-million-dollar deals from Amazon, Apple or Spotify — the industry isn’t as imperiled as it seems. And yet, Spotify’s shadow looms so large over the podcasting industry that it’s impossible for its failures not to reverberate.
In 2021, a year that saw venture capital flowing like champagne at a Gatsby party, Spotify CEO Daniel Ek told Forbes that he wanted his company to be like the Instagram or TikTok of audio.
“Everyone underestimates audio. It should be a multi-hundred-billion-dollar industry,” Ek said at the time. “Audio is ours to win.”
In the last handful of years, we watched as Spotify acquired too many podcasting companies to count — Gimlet, The Ringer, Anchor, Parcast, Megaphone — and then courted big names from Joe Rogan, to Alex Cooper, to Prince Harry with eight- and nine-figure deals. The company dumped over a billion dollars into its efforts to corner podcasting but now has canceled over a dozen shows from the studios it spent so many hundreds of millions to acquire, like Parcast and Gimlet, which have since been combined into one entity and decimated.
“In hindsight, I was too ambitious in investing ahead of our revenue growth,” Ek said after Spotify laid off 600 people in January.
After acquiring Gimlet and Parcast, Spotify made most of the networks’ shows exclusive to the Spotify platform. In theory, this decision would force the listeners of these popular shows to download Spotify in order to keep listening every week — and, hopefully, some of those listeners would convert to paid subscribers. But, according to the Gimlet and Parcast unions, this strategy backfired. Some shows lost more than three-quarters of their audiences after being converted to Spotify exclusives.
“Spotify told show teams that their podcasts were being canceled because of low numbers,” said a joint statement by the Gimlet and Parcast unions, posted after a round of layoffs in October 2022. “But decisions Spotify leadership made directly contributed to these low numbers.”
This isn’t necessarily a bad thing. During the funding boom of 2021, my inbox was flooded with pitches from creator economy startups seeking press about their latest funding rounds. Some of these companies were exciting, but many of them confused me — as a creator myself, I couldn’t imagine myself or my friends in the industry using many of these products. As SignalFire partner Josh Constine told me earlier this year, “Creators aren’t sophisticated enterprise software buyers, nor do they have software integration teams.” In other words, VCs had thrown money at companies that didn’t actually solve any problems for creators’ businesses. So, I wasn’t surprised that when market conditions tightened, the companies that seemed to only want to capitalize on the creator economy hype were no longer being funded.
“A media company needs to have the goal of making enough money for it to survive,” Silver told TechCrunch. This might seem intuitive — surely, a business should try to turn a profit. But that’s not how the world of venture-funded startups works. Spotify, for example, has only reported quarterly profits a handful of times, because its business has prioritized continual growth over returns. The company is not at all unique in that way.
“It’s critical for companies to ‘read the market,’ and right now, the market values efficient growth and doing more with less instead of maximum growth with easy capital,” Creative Juice founder Sima Gandhi told TechCrunch this summer.
This “maximum growth” mindset has poisoned venture-backed digital media companies like Buzzfeed, which descended from a shining star to an IPO embarrassment. The “middle class” of podcasters can’t rely on Spotify, and other media workers can’t rely on failing media conglomerates like G/O Media and Vice anymore. Over the last few years, worker-owned media outlets like Defector, Aftermath and 404 Media have begun cropping up, often founded and staffed by journalists who had been repeatedly laid off from mismanaged media companies. Now the podcasting industry is facing the same reckoning as Spotify’s losses prove that growth can’t take priority over sustainability. Already, podcast studio Maximum Fun has adopted a worker-ownership co-op model, and as podcasters continue to lose trust in big corporations like Spotify, we’ll see this trend continue.
“Spotify is not all of podcasting, although they act as if they are, and make choices as if they’re the only one in the room,” Silver said. “Podcasting is not dead.”
Sarah Perez @sarahpereztc / 8:00 AM PST•December 7, 2023
Image Credits: Guillaume Payen/SOPA Images/LightRocket / Getty Images
In September, Google announced it would shut down its standalone podcasts app, Google Podcasts, sometime next year. Now that the end of 2023 is nearing, the company is today launching a migration tool that will allow U.S. users to shift their existing podcast subscriptions over to YouTube Music, which will be Google’s new home for podcasts.
Users will have plenty of time to export their subscriptions as the official discontinuation of Google Podcasts won’t take place until April 2024. While the current tool only supports U.S. users, Google says it will become available to other markets soon.
The company additionally shared the timeline for the Google Podcasts app’s shutdown. It notes that U.S users will be able to listen to their podcasts in the app through March 2024. And although the app is being discontinued in April, users will still be able to migrate or export subscriptions through July 2024.
The new migration tool will appear in the app in the weeks ahead, through a banner at the top of the screen. Step-by-step instructions will also be documented on Google’s support site. For those who don’t want to move to YouTube Music, an option to export subscriptions to an OPML file will also be provided. This file can be uploaded into any other third-party podcast app that also supports uploads.
Google has been working to make YouTube more of a destination for podcasts for some time, launching a dedicated podcasts homepage last year, and announcing plans this February to bring podcasts to YouTube Music — the company’s Spotify and Apple Music rival.
The change will consolidate Google’s efforts in the audio streaming market by allowing it to combine the listenership that’s today spread across multiple apps. The company had already shuttered Google Play Music years ago, but it took longer to address podcasts.
by Jason Lemkin | Blog Posts
So things aren’t all “bad” in SaaS. Many Cloud leaders stock prices are way, way up in 2023, the Cloud platform leaders have re-accelerated, and leaders like Shopify are having close-to-record years.
Now it sort of has to be that way. So, so many SaaS startups got funded in the Boom, and they just can’t all make it. In fact, Craft’s latest data from the presentation below has seen as many as 80% of Seed startups fail to raise a Series A.
So we’re working through this. But Be Kind. The shutdowns in 2023 have been at a record pace, and it’s only going to continue in 2024. The latest data from Pilot (see below) suggests even more startups will likely run out of runway in 2024. Hopefully, we’ll be through most of it by the end of 2024.
by Jason Lemkin | Blog Posts, Fundraising
SaaS products and services like Pilot track the finances of 1,000s of SaaS and other startup so they’re an interesting source of hard data.
What does Pilot’s latest data say? Something that’s both not surprising but also pretty impactful: 57% of venture-backed startups will have to go “back to market” in 2024 to raise more capital. And 38% have 12 or less months of runway left. Many have already raised a bridge round. And realistically, most won’t have the metrics to pull off another round.
That’s how it works. VCs don’t give startups 10 years of capital. They typically give them just enough to see if it will work, and the startup grows and scales to the next stage. That’s typically been 18 months historically.
At a practical level though, the headlines in 2024 may actually look much worse than 2023 for startup failures. So many were able to cut the burn and stretch their cash through 2023. But you can only cut so much. Ultimately, you also have to grow again to raise more venture capital.
Folks that raised in the Go-Go Times of 2021 in many cases have been able to stretch their cash through 2023. But many will find 2024, those stretches have stretched as far as they can.
SaaS and Cloud growth overall will remain strong. Shopify, Datadog, Crowdstrike, Google Cloud-Azure-AWS, Snowflake, etc. may well put up not just strong numbers, but even stronger than 2023. In fact, Gartner predicts enterprise software spend will cross $1 Trillion Dollars (!) for the first time in 2024!
But with so, so many startups funded in 2021 … we likely will also see a record number shut down in 2024. :(. That’s what this Pilot data also reflects.
So let’s be empathetic to those that wind down in 2024. But also focus on the prize at the same time — record Cloud spend. Carpe Diem.
Will Shanklin, Contributing Reporter’ Fri, Dec 8, 2023
OpenAI’s recent drama hasn’t only caught UK regulators’ attention. Bloomberg reported Friday that the Federal Trade Commission (FTC) is looking into Microsoft’s investment in the Sam Altman-led company and whether it violates US antitrust laws. FTC Chair Lina Khan wrote in a New York Times op-ed earlier this year that “the expanding adoption of AI risks further locking in the market dominance of large incumbent technology firms.”
Bloomberg’s report stresses that the FTC inquiry is preliminary, and the agency hasn’t opened a formal investigation. But Khan and company are reportedly “analyzing the situation and assessing what its options are.” One complicating factor for regulation is that OpenAI is a non-profit, and transactions involving non-corporate entities aren’t required by law to be reported.
In addition, Microsoft’s $13 billion investment doesn’t technically give it control over OpenAI in the eyes of the law, another factor in determining what action a governmental agency might be able to take. However, the recent ousting and re-hiring of Altman — and the integral role Microsoft played in reverting those chess pieces to its preferred positions — suggests the lack of control over the nonprofit is more a technicality than the relationship’s underlying essence.
The UK’s Competition and Markets Authority (CMA) wrote earlier today that it’s considering investigating the relationship between AI’s two dominant players. It said it’s weighing “recent developments,” referring obliquely to the Altman-Microsoft drama. “The CMA will review whether the partnership has resulted in an acquisition of control — that is, where it results in one party having material influence, de facto control or more than 50% of the voting rights over another entity,” the CMA wrote in its news release.
Khan, also challenging Microsoft’s $69 billion Activision Blizzard acquisition, has previously sounded the alarm about the need for AI regulations.
“As these technologies evolve, we are committed to doing our part to uphold America’s longstanding tradition of maintaining the open, fair and competitive markets that have underpinned both breakthrough innovations and our nation’s economic success — without tolerating business models or practices involving the mass exploitation of their users,” the youngest-ever FTC chair wrote in May. “Although these tools are novel, they are not exempt from existing rules, and the F.T.C. will vigorously enforce the laws we are charged with administering, even in this new market.”
Amazon filed a motion on Friday in the Western Washington district court asking a judge to dismiss the Federal Trade Commission’s (FTC) antitrust lawsuit against it. The FTC along with 17 state attorneys general sued Amazon in September, alleging the company uses monopolistic practices that are unfair to both its competitors and consumers. Amazon is now arguing that the FTC did not provide evidence that its practices have driven up prices or harmed consumers, according to Bloomberg.
The FTC’s lawsuit claims Amazon uses illegal tactics to crush its competition — like punishing sellers who list their products for better prices elsewhere by burying them in search results, and coercing sellers into using Amazon’s own fulfillment service by tying it to Prime eligibility. It also accuses Amazon of inflating prices from 2016-2018 using an algorithmic tool codenamed Project Nessie. These increases added up to more than $1 billion, according to the suit.
In Amazon’s motion for dismissal, per AP, Amazon said it’s only engaging in “common retail practices” that “benefit consumers and are the essence of competition.” Amazon attorney Heidi Hubbard wrote that the suit “implausibly, and illogically, assumes that Amazon’s efforts to keep featured prices low on Amazon somehow raised consumer prices across the whole economy,” according to Bloomberg.
Devin Coldewey @techcrunch / 2:06 PM PST•December 4, 2023
Image Credits: Alibaba Group
As if still-image deepfakes aren’t bad enough, we may soon have to contend with generated videos of anyone who dares to put a photo of themselves online: with Animate Anyone, bad actors can puppeteer people better than ever.
The new generative video technique was developed by researchers at Alibaba Group’s Institute for Intelligent Computing. It’s a big step forward from previous image-to-video systems like DisCo and DreamPose, which were impressive all the way back in summer but are now ancient history.
What Animate Anyone can do is not by any means unprecedented, but has passed that difficult space between “janky academic experiment” and “good enough if you don’t look closely.” As we all know, the next stage is just plain “good enough,” where people won’t even bother looking closely because they assume it’s real. That’s where still images and text conversation are currently, wreaking havoc on our sense of reality.
Image-to-video models like this one start by extracting details, like facial feature, patterns and pose, from a reference image like a fashion photo of a model wearing a dress for sale. Then a series of images is created where those details are mapped onto very slightly different poses, which can be motion-captured or themselves extracted from another video.
Previous models showed that this was possible to do, but there were lots of issues. Hallucination was a big problem, as the model has to invent plausible details like how a sleeve or hair might move when a person turns. This leads to a lot of really weird imagery, making the resulting video far from convincing. But the possibility remained, and Animate Anyone is much improved, though still far from perfect.
The technical specifics of the new model are beyond most, but the paper emphasizes a new intermediate step that “enables the model to comprehensively learn the relationship with the reference image in a consistent feature space, which significantly contributes to the improvement of appearance details preservation.” By improving the retention of basic and fine details, generated images down the line have a stronger ground truth to work with and turn out a lot better.
Image Credits: Alibaba Group
They show off their results in a few contexts. Fashion models take on arbitrary poses without deforming or the clothing losing its pattern. A 2D anime figure comes to life and dances convincingly. Lionel Messi makes a few generic movements.
They’re far from perfect — especially about the eyes and hands, which pose particular trouble for generative models. And the poses that are best represented are those closest to the original; if the person turns around, for instance, the model struggles to keep up. But it’s a huge leap over the previous state of the art, which produced way more artifacts or completely lost important details like the color of a person’s hair or their clothing.
It’s unnerving thinking that given a single good-quality image of you, a malicious actor (or producer) could make you do just about anything, and combined with facial animation and voice capture tech, they could also make you express anything at the same time. For now, the tech is too complex and buggy for general use, but things don’t tend to stay that way for long in the AI world.
At least the team isn’t unleashing the code into the world just yet. Though they have a GitHub page, the developers write: “we are actively working on preparing the demo and code for public release. Although we cannot commit to a specific release date at this very moment, please be certain that the intention to provide access to both the demo and our source code is firm.”
Will all hell break loose when the internet is suddenly flooded with dancefakes? We’ll find out, and probably sooner than we would like.
Share Dialog
Share Dialog
No activity yet