Dreamer which launched just a month ago by Hugo Barra and David Singleton got execu-hired by Meta and joined Meta's Superintelligence Labs. I'm also very piqued by Littlebird with its 'agent-on-a-shoulder' approach via 'screen reading'.
Whilst Anthropic definitely impacted its fans by tightening Claude's usage during peak hours, both OpenAI and Zhipu responding with increased rate limits and in the case of Zhipu launch of GLM-5.1 exclusively for now via its coding plan and showing via its benchmarks much improved numbers relative to GLM-5 and even approaching Opus 4.6
I had some issues using the table feature of Paragraph whilst creating this week's newsletter so layout will be different compared to previous weeks
Readers who think others in their family, friends and acquaintances who are curious in knowing more about rapidly evolving AI tools/services/use cases and 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 22nd Mar 2026 to Saturday 28th Mar 2026
Supermemory introduced ASMR—Agentic Search and Memory Retrieval—an experimental in-memory system using parallel AI agents that scored 98.6% on LongMemEval-s, crushing prior results like their own 85% and rivals' lower marks. It skips vector databases entirely, with observer agents extracting knowledge from long chat histories and specialized search agents handling queries on facts, context, and timelines. Founder Dhravya Shah shared the breakthrough, planning to open-source the code soon, while excitement builds alongside concerns over high compute costs for real-world scaling.
OpenAI Releases GPT-5.4 with Frontend Design Skills and Prompting Playbook
OpenAI released the GPT-5.4 model featuring new frontend design capabilities, including skills like frontend-skill and design-exploration. The release includes a prompting playbook to improve frontend outputs. An OpenAI blog post highlights better frontend results using low or medium reasoning settings.
Practical techniques for steering GPT-5.4 toward polished, production-ready frontend designs.
Meta acqui-hires Hugo Barra's/David Singleton's dreamer.com
Littlebird definitely has my 'must-try-out' interest given that it has S-Tier folks such as Gokul Rajaram and Lenny amongst its angel investors
Littlebird raises $11M for its AI-assisted ‘recall’ tool that reads your computer screen
Currently MacOS only, According to a message to me from founder Alexander Green, it's coming to Android soon but this approach (screenreading) is not possible on iOS given Apple's platform limitations.
Littlebird is building an AI that reads your screen in real time to capture context, answer questions, and automate tasks, without relying on screenshots.
via Alex
Famous german lawyer Dr. Jun talking about automated copyright infringements when using AI coding. His argument: AI code without human process & architecture work is not a greenfield implementation and thus a copyright infringement.
Especially because models are known to ignore licenses (e.g. GPL) and reissue the verbatim code under MIT
By the way this is the Chardet case referenced in the talk
Alarm bells are ringing in the open source community, but commercial licensing is also at risk
via Robby Yung
Introducing the MiniMax Token Plan: First All-Modality API Subscription
Flat-rate API access to MiniMax's leading text, speech, music, video, and image models.
Every major image model for the last three years has worked the same way: start with static, remove noise until a picture appears. Diffusion. Midjourney, Stable Diffusion, DALL-E, Imagen. All variations on the same idea.
Uni-1 doesn’t use diffusion. It generates images token by token, the same autoregressive architecture that powers GPT and Claude for text. One model that processes the prompt and produces the image in a single pass.
via Yat Siu
not sure if this is the right forum but for people living in HK this remains a concern
Reply by Yusuf
Whilst its preferrable that group members don't delve too much into the realm of geopolitics since that will very likely open up a Pandora's box of opinion and in the end its the right of an organisation/nation-state to decide whom they want to do business with. I've mentioned the challenges faced by consumers and developers in Hong Kong in the Notion page of mine which is part of my AI/ML Resource Hub
https://yusuf-goolamabbas-53.notion.site/AI-services-that-I-subscribe-to-aee845c3adf0497a8563651d72742624?source=copy_link
via Daniel Luca
I just published this
via Robby Yung
Anthropic Tightens Claude AI Limits During Peak Hours Amid Surging Demand
Anthropic has adjusted its Claude AI to deplete session limits quicker on weekdays from 5 a.m. to 11 a.m. PT, affecting free, Pro, and Max users while keeping weekly totals the same. Thariq from Anthropic explained the change distributes usage more evenly and noted efficiency gains to soften the blow, though about 7% of users, especially Pro coders, face new hurdles with token-heavy work like extended coding sessions. OpenAI responded by promoting its Codex tool with temporary higher limits, as power users vent about quick caps and some even cancel plans.
Cohere Transcribe Tops Open ASR Leaderboard with New Speech Model
The 2-billion-parameter Cohere Transcribe, achieved a top average word error rate of 5.42% across English benchmarks, beating Whisper Large v3's 7.44% and others. It excels in human preference tests at 61% win rate for handling accents, noise, and multiple speakers, while processing audio up to three times faster. Supporting 14 languages from English to Korean under an Apache 2.0 license, it runs on consumer GPUs and powers enterprise tools—coinciding with Mistral's Voxtral TTS launch for a big step in open voice AI.
GLM-5.1 available with benchmarks from Zhipu showing much improved coding performance relative to GLM-5 but currently only via Coding Plan.
From the ZAI Discord
The current model is still in the preview version. We will continue to optimize it based on everyone's feedback, so we haven't open-sourced it for now. We plan to open-source the model and launch the official full version around April 6th or April 7th.
Below is my personal website which aggregates links to many of my socials as well as the various 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 Google Nano Banana 2 model within the Freepik tool via the following prompt
The lost city of Atlantis during a foggy morning, with a moderately chaotic ocean. Mist swirls around the tall, ancient architectures. Various sealife, including curious sharks, schools of fish, and diverse coral, are partially obscured. Octane render, Unreal Engine 5, 8k, detailed.
Dreamer which launched just a month ago by Hugo Barra and David Singleton got execu-hired by Meta and joined Meta's Superintelligence Labs. I'm also very piqued by Littlebird with its 'agent-on-a-shoulder' approach via 'screen reading'.
Whilst Anthropic definitely impacted its fans by tightening Claude's usage during peak hours, both OpenAI and Zhipu responding with increased rate limits and in the case of Zhipu launch of GLM-5.1 exclusively for now via its coding plan and showing via its benchmarks much improved numbers relative to GLM-5 and even approaching Opus 4.6
I had some issues using the table feature of Paragraph whilst creating this week's newsletter so layout will be different compared to previous weeks
Readers who think others in their family, friends and acquaintances who are curious in knowing more about rapidly evolving AI tools/services/use cases and 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 22nd Mar 2026 to Saturday 28th Mar 2026
Supermemory introduced ASMR—Agentic Search and Memory Retrieval—an experimental in-memory system using parallel AI agents that scored 98.6% on LongMemEval-s, crushing prior results like their own 85% and rivals' lower marks. It skips vector databases entirely, with observer agents extracting knowledge from long chat histories and specialized search agents handling queries on facts, context, and timelines. Founder Dhravya Shah shared the breakthrough, planning to open-source the code soon, while excitement builds alongside concerns over high compute costs for real-world scaling.
OpenAI Releases GPT-5.4 with Frontend Design Skills and Prompting Playbook
OpenAI released the GPT-5.4 model featuring new frontend design capabilities, including skills like frontend-skill and design-exploration. The release includes a prompting playbook to improve frontend outputs. An OpenAI blog post highlights better frontend results using low or medium reasoning settings.
Practical techniques for steering GPT-5.4 toward polished, production-ready frontend designs.
Meta acqui-hires Hugo Barra's/David Singleton's dreamer.com
Littlebird definitely has my 'must-try-out' interest given that it has S-Tier folks such as Gokul Rajaram and Lenny amongst its angel investors
Littlebird raises $11M for its AI-assisted ‘recall’ tool that reads your computer screen
Currently MacOS only, According to a message to me from founder Alexander Green, it's coming to Android soon but this approach (screenreading) is not possible on iOS given Apple's platform limitations.
Littlebird is building an AI that reads your screen in real time to capture context, answer questions, and automate tasks, without relying on screenshots.
via Alex
Famous german lawyer Dr. Jun talking about automated copyright infringements when using AI coding. His argument: AI code without human process & architecture work is not a greenfield implementation and thus a copyright infringement.
Especially because models are known to ignore licenses (e.g. GPL) and reissue the verbatim code under MIT
By the way this is the Chardet case referenced in the talk
Alarm bells are ringing in the open source community, but commercial licensing is also at risk
via Robby Yung
Introducing the MiniMax Token Plan: First All-Modality API Subscription
Flat-rate API access to MiniMax's leading text, speech, music, video, and image models.
Every major image model for the last three years has worked the same way: start with static, remove noise until a picture appears. Diffusion. Midjourney, Stable Diffusion, DALL-E, Imagen. All variations on the same idea.
Uni-1 doesn’t use diffusion. It generates images token by token, the same autoregressive architecture that powers GPT and Claude for text. One model that processes the prompt and produces the image in a single pass.
via Yat Siu
not sure if this is the right forum but for people living in HK this remains a concern
Reply by Yusuf
Whilst its preferrable that group members don't delve too much into the realm of geopolitics since that will very likely open up a Pandora's box of opinion and in the end its the right of an organisation/nation-state to decide whom they want to do business with. I've mentioned the challenges faced by consumers and developers in Hong Kong in the Notion page of mine which is part of my AI/ML Resource Hub
https://yusuf-goolamabbas-53.notion.site/AI-services-that-I-subscribe-to-aee845c3adf0497a8563651d72742624?source=copy_link
via Daniel Luca
I just published this
via Robby Yung
Anthropic Tightens Claude AI Limits During Peak Hours Amid Surging Demand
Anthropic has adjusted its Claude AI to deplete session limits quicker on weekdays from 5 a.m. to 11 a.m. PT, affecting free, Pro, and Max users while keeping weekly totals the same. Thariq from Anthropic explained the change distributes usage more evenly and noted efficiency gains to soften the blow, though about 7% of users, especially Pro coders, face new hurdles with token-heavy work like extended coding sessions. OpenAI responded by promoting its Codex tool with temporary higher limits, as power users vent about quick caps and some even cancel plans.
Cohere Transcribe Tops Open ASR Leaderboard with New Speech Model
The 2-billion-parameter Cohere Transcribe, achieved a top average word error rate of 5.42% across English benchmarks, beating Whisper Large v3's 7.44% and others. It excels in human preference tests at 61% win rate for handling accents, noise, and multiple speakers, while processing audio up to three times faster. Supporting 14 languages from English to Korean under an Apache 2.0 license, it runs on consumer GPUs and powers enterprise tools—coinciding with Mistral's Voxtral TTS launch for a big step in open voice AI.
GLM-5.1 available with benchmarks from Zhipu showing much improved coding performance relative to GLM-5 but currently only via Coding Plan.
From the ZAI Discord
The current model is still in the preview version. We will continue to optimize it based on everyone's feedback, so we haven't open-sourced it for now. We plan to open-source the model and launch the official full version around April 6th or April 7th.
Below is my personal website which aggregates links to many of my socials as well as the various 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 Google Nano Banana 2 model within the Freepik tool via the following prompt
The lost city of Atlantis during a foggy morning, with a moderately chaotic ocean. Mist swirls around the tall, ancient architectures. Various sealife, including curious sharks, schools of fish, and diverse coral, are partially obscured. Octane render, Unreal Engine 5, 8k, detailed.
This Week in All Things AIThis Week in All Things AI - Inaugural Edition
Greetings friends, Whilst you may have subscribed to my low-volume newsletter which focused on informing what had changed in the various public Notion pages that I curate, I decided to experiment with a new publication/newsletter which I hope to send on a weekly basis that aggregates what was posted the previous week on the 'All Things AI' telegram group Some of the reasons that I want to do this experimentThere are folks who have mentioned to me that they are not on Telegram and don't wish t...
This Week in All Things AIThis Week in All Things AI - Week 35-2025
Sunday 24th August 2025 and Saturday 30th August 2025
This Week in All Things AIThis Week in All Things AI - Week 39-2025
Sunday 21st September 2025 to Saturday 27th September 2025
This Week in All Things AIThis Week in All Things AI - Inaugural Edition
Greetings friends, Whilst you may have subscribed to my low-volume newsletter which focused on informing what had changed in the various public Notion pages that I curate, I decided to experiment with a new publication/newsletter which I hope to send on a weekly basis that aggregates what was posted the previous week on the 'All Things AI' telegram group Some of the reasons that I want to do this experimentThere are folks who have mentioned to me that they are not on Telegram and don't wish t...
This Week in All Things AIThis Week in All Things AI - Week 35-2025
Sunday 24th August 2025 and Saturday 30th August 2025
This Week in All Things AIThis Week in All Things AI - Week 39-2025
Sunday 21st September 2025 to Saturday 27th September 2025
Weekly newsletter which reposts content from the 'All Things AI' telegram group curated by Yusuf Goolamabbas
Weekly newsletter which reposts content from the 'All Things AI' telegram group curated by Yusuf Goolamabbas
Share Dialog
Share Dialog
Subscribe to This Week in All Things AI
Subscribe to This Week in All Things AI
>800 subscribers
>800 subscribers
No activity yet