Wishing everyone and their loved ones a very Happy New Year
Steve Yegge's 37 minute Youtube video linked below is what I've been sending to many engineers alongside Andrej Karpathy's and Boris Cherny's (creator of Claude Code) tweets to get them to understand the shakeup that is coming in AI assisted coding
Deepseek's paper which gives a hint at what is likely to come with Deepseek v4 as well how Chinese open weights/open source models might evolve in 2026 is exciting many people
Whilst unrelated to AI per-se, I am sharing 3 links below from some of my other public Notion pages that I curate which I thought might be of interest to readers of this newsletter. Feel free to share these links to others
Public: Runnin' Down A Dream: How to Succeed and Thrive in a Career You Love by Bill Gurley | Notion
The following links summarise a presentation by Bill Gurley [famous venture capitalist] at the McCombs School at the University of Texas on September 14th, 2018 titled "Runnin' Down a Dream: How to Succeed and Thrive in a Career You Love."
Public: Bill Gurley talk at All-in-Summit on Regulatory Capture | Notion
In this session from the All-In Summit, venture capitalist Bill Gurley outlines the dangers of regulatory capture, where established industries manipulate government rules for their own benefit.
Public: Don't Follow Your Passion | Notion
This 17 minutes speech from Ben Horowitz of the famous venture capital firm Andressen Horowtiz aka A16Z of the 2015 commencement address at Columbia University has a click-baity title but in my humble opinion is worth watching
Readers who think others in their family, friends and acquaintances who have a curiosity in knowing more about rapidly evolving AI tools/services/use cases would benefit from being subscribers of this weekly newsletter are encouraged to share this publication link to them and invite them to subscribe
The following messages were posted on the 'All Things AI' Telegram group from Sunday 28th Dec 2025 to Saturday 3rd Jan 2026
In this 37 min interview, tech veteran Steve Yegge introduces vibe coding, a paradigm shift where developers move away from traditional manual coding to orchestrating autonomous agents. He argues that the industry is entering a "factory farming" era of software, where high-performance engineers use AI to achieve 10x productivity gains that leave traditionalists behind.
Yegge suggests that while manual IDE use is becoming obsolete, engineers must still possess deep architectural knowledge to guide these AI systems effectively.
He highlights significant challenges like merging massive AI-generated codebases and the steep learning curve required to truly trust and master agentic workflows.
Ultimately, the discussion frames AI engineering not just as a new toolset, but as a fundamental reimagining of the software lifecycle that favors speed and high-level strategy over syntax.
Meta agrees to acquire Singapore-based startup Manus, which makes an AI agent it sells to SMBs; Manus said in December its annual revenue run rate was $125M+
Meta says it will continue to operate and sell the Manus service, and Manus' talent will join Meta to deliver agents across Meta products, including Meta AI
Full story on our blog:
Manus Joins Meta for Next Era of Innovation
Manus is joining Meta, and we’ll continue delivering our current services while accelerating product improvements to bring more powerful, reliable general AI agent capabilities to more users and...
The Manus team in Singapore are world class at exploring the capability overhang of today’s models to scaffold powerful agents.
Looking forward to working with you, @Red_Xiao_!
Manus Joins Meta: Accelerating AI Innovation for Businesses
We are excited to announce that Manus is joining Meta to bring a leading agent to billions of people and unlock opportunities for businesses across our products.
1 | Lovable’s Head of Growth on Growth Playbooks in the AI Era Recently I came across two videos on Youtube with a total viewing time of over 2 hours which had Lovable's Head of Growth Elena Verna arguing that traditional marketing strategies are failing, requiring companies to shift their focus toward product-led distribution to survive. I created this Notion page which collates the videos and provides some summaries via Google's NotebookLM (feel free to use any of your other AI tools for your analysis of the videos) I hope that some of you find this content useful and interesting |
2 | Simon Willinson's 2025 takeaways in LLMs: reasoning as a signature feature, coding agents were useful, subscriptions hit $200/month, and Chinese open-weight models impressed |
3 | Came across this post on X which summarized all that was released on Replit throughout 2025 |
4 | DeepSeek Unveils mHC for Stable Hyper-Connections in Large AI Models The team, led by Zhenda Xie, Yixuan Wei, and Huanqi Cao with founder Wenfeng Liang, introduced mHC—Manifold-Constrained Hyper-Connections. It projects hyper-connection matrices onto doubly stochastic manifolds to preserve signal balance, backed by custom kernels, memory tweaks, and efficiency upgrades. Observers praised the full-stack work—from math proofs to GPU optimizations—as a major step for stable, wide-stream training in massive models.  DeepSeek just dropped a paper that quietly exposes why modern neural networks get unstable as they scale.It’s called mHC: Manifold-Constrained Hyper-Connections, and the core idea is deceptively simple: Neural networks keep breaking their own geometry. Here’s what that 
DeepSeek ended the year with a bang and dropped another paper. A change in transformer design. What it means (simplified)?
1. Better Memory (Feature Reuse) Old way: If Layer 10 needs information from Layer 1, Layer 1 has to pass it to 2, then to 3, then to 4... all the way to 
arxiv.org
mHC: Manifold-Constrained Hyper-ConnectionsRecently, studies exemplified by Hyper-Connections (HC) have extended the ubiquitous residual connection paradigm established over the past decade by expanding the residual stream width and...
DeepSeek drops new paper: mHC (Manifold-Constrained Hyper-Connections). According to a DeepSeek researcher, "the two biggest architectural innovations in 2025 are 1) Muon and 2) Hyper-Connections." So what does mHC bring to the table?
Residual connections have been the backbone 
for all the ppl that don't understand shit about this(like me), summary:
> future models will be better at complex reasoning and math without needing to just be "bigger" > DeepSeek is ditching the standard AI structure for a new "Hyper-Connected" design that handles more
ALERT, NEW YEAR GIFT FROM DEEPSEEK
mHC: Manifold-Constrained Hyper-Connections it's a pretty crazy fundamental result! They show stable hyper-connection training. This leth them *scale residual stream width*, with minor compute&memory overhead This is a *huge model smell* recipe.  |
Google's NotebookLM Data Tables feature turns scattered research into organized, structured tables you can export directly to Google Sheets. There are many other features of NotebookLM, I encourage group members to check out NotebookLM at their convenience and ideally a paid version of it
NotebookLM can now turn your messy research into structured Data Tables
If you use NotebookLM, you know the magic: you dump in a dozen PDFs, transcripts, and notes, and it instantly becomes an expert on your data. But until now, getting that data out of NotebookLM in a structured way was a bit of a manual chore. That changes today.
NotebookLM's Last Gift: Data Tables Changes Everything
New NotebookLM Data Tables functionality automatically converts scattered information into structured insights. Learn the progressive synthesis strategy.
1 | Google Engineer: Claude Code Built Year-Long Project in One Hour Jaana Dogan, who works on Google's Gemini API and has shaped tools like Google Drive and Spanner, shared that Claude Code created a distributed agent orchestrator in her terminal, handling scheduling, interactions, and consistency across networks. She emphasized the output isn't perfect but shows AI's real power, especially after internal Google debates dragged on her team's year-long effort. Dogan urged experts to test it themselves on familiar domains, praising the tool while noting her prompt was just three paragraphs for a toy version built on existing ideas. Also sharing a tweet from Boris Cherny, creator of Claude Code on his setup
I'm not joking and this isn't funny. We have been trying to build distributed agent orchestrators at Google since last year. There are various options, not everyone is aligned... I gave Claude Code a description of the problem, it generated what we built last year in an hour.
I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit.
My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to |
2 | Came across Zero Lu, a curator of excellent prompts for Gemini and Nano-Banana who also has a low-volume newsletter. Hope this is useful to members of this group https://zerolu.substack.com/p/hello-there |
Below is my personal website which aggregates links to many of my socials as well as content and community that I curate. Feel free to share this link to others who you think may find this content/community useful to them
The cover image of this newsletter via generated via the Seedream 4.5 model within the Freepik tool via the following prompt
hyperrealistic cinematic photo, a magical forest where trees are made of smooth, polished geode stone. The bark is a dark basalt that cracks open to reveal vast, shimmering cavities of giant amethyst and clear quartz crystals. Sunlight filters through the canopy and refracts into thousands of tiny, sharp rainbows that dance on the mossy forest floor. The air is crisp and clear, with a few particles of dust illuminated in the light beams. Shot on Arri Alexa 65, anamorphic lens, epic scale, breathtaking detail.