Everything in AI worth keeping in context.
Hugging Face published a write-up on scaling pretraining for Indic languages to 2.8 trillion tokens, along with an open data generation pipeline. Few further details are available beyond the title.
Calorie-tracking app Cal AI generated over $30 million in revenue in two years, largely through a large network of paid influencers, before MyFitnessPal acquired it in March for an undisclosed sum.
Open sourceBriefing
Hugging Face, The Verge +11 min read
OpenAI launched the Rosalind Academic Plan, giving eligible academic labs sponsored access to GPT-Rosalind and tools for life sciences research.
Google is rolling out Universal Cart in the US across the Gemini app and Google Search, letting users save products from multiple retailers in one cart. It was previewed at I/O in May.
Nick Cammarata claims that AI agents disprove polysemanticity and substantially improve interpretability and alignment, according to Latent Space's AINews roundup. Details of the evidence are not given in the snippet.
Nick Cammarata claims that AI agents disprove polysemanticity and substantially improve interpretability and alignment, according to Latent Space's AINews roundup. Details of the evidence are not given in the snippet.
Hugging Face published a write-up on scaling pretraining for Indic languages to 2.8 trillion tokens, along with an open data generation pipeline. Few further details are available beyond the title.
Calorie-tracking app Cal AI generated over $30 million in revenue in two years, largely through a large network of paid influencers, before MyFitnessPal acquired it in March for an undisclosed sum.
ChatGPT's Chrome extension can now use your open tabs when you tag Chrome in the web or desktop app, bringing computer use to Chrome users; the Atlas browser is gone.
Sega pulled Total War: Shogun 2 - Complete Edition from sale after a launch rated Overwhelmingly Negative on Steam, with players blaming bugs and apparently AI-botched upscaled artwork.
2026 films such as Aaron Sorkin's The Social Reckoning and You Can See Everything take a critical view of Silicon Valley figures, reflecting a broader tech backlash.
Trump posted White House renovation photos on Truth Social, one of which Google's AI detector flagged as made or edited with OpenAI, with inconsistent step counts between before and after images.
Granite Switch 4.1 is a release that serves chat, decisions and speech from a single endpoint.
Amazon devices chief Panos Panay said at TechCrunch Disrupt that ambient AI devices working in the background and anticipating needs could replace the need to use apps.
OpenAI launched the Rosalind Academic Plan, giving eligible academic labs sponsored access to GPT-Rosalind and tools for life sciences research.
Bluesky expanded its experimental AI-powered Attie app with "social vibe coding", letting users build interactive websites, games and quizzes that can tap into the social network's audience.
Current Affairs explains in an essay why it is banning AI from its publication.
Google DeepMind describes using Google Antigravity for Science to speed up the search for novel antimicrobial peptides.
Goodfire argues in a post titled "Neural networks are clean" about the interpretability of neural networks; only the title is available.
Microsoft Research unveiled Magnetar, a system for zero-knowledge proof-based computing that supports standard programming languages, claims major speed gains, and offers quantum-resistant security.
The FTC is reportedly probing Anthropic and OpenAI after Trump dismissed AI fears as a "hoax"; the agency may fine AI firms comparably to companies that delay data breach reports.
Rivian plans to bring AI agents to its cars, with details expected at its AI & Autonomy Day event in December.
The Simons Foundation launched a $15M effort to help mathematicians adapt to the AI era.
Signalscorps analyzed 217 million web responses to measure prompt injection attempts placed in HTTP headers.
A GNOME blog post describes using AI to reverse engineer hardware.
Microsoft and Nvidia launched the first RTX Spark laptops, which use Arm-based chips with a powerful GPU and target AI PCs. Windows is being optimized for the platform.
A fake Instagram account with over a million followers and a verified badge posted an AI-generated image of actress Mia Goth, fooling many users. The case illustrates how celebrity deepfakes are spreading even to low-profile stars.
Google is rolling out Universal Cart in the US across the Gemini app and Google Search, letting users save products from multiple retailers in one cart. It was previewed at I/O in May.
OpenAI is promoting its Study Mode for teachers in Brazilian Portuguese, tied to Teachers' Day, to support active, step-by-step learning.
Amazon announced the $74.99 Bee Frontier and the coming $99.99 Bee Eclipse, always-listening AI wearables that summarize conversations and now help with shopping, along with a new subscription.
The Verge found that Eiliko, a cute AI companion robot toy from Energize Lab, sends conversations to Chinese servers where they can be easily accessed or hijacked, and the company seems unconcerned.
Modal raised a Series D at a $16B valuation with $1B in annual recurring revenue, with usage and valuation both tripling in five months.
Apertus 1.5, the open model, has release notes published on Hugging Face.
A Substack essay offers a personal take on AI losing its appeal.
Figma announced features to define how AI agents work on its platform and released its code layers feature in open beta.
Granite Switch 4.1 is a release that serves chat, decisions and speech from a single endpoint.
ModelsBriefing6 sources
Gemini 4 Argon is available via API at an introductory $2/$10 per million tokens, and to Pro and Ultra subscribers in the Gemini app. Antigravity access is limited to Ultra, and Google says it ranks first on the Vals Index.
Google AI, Latent Space +51 min read
Goodfire argues in a post titled "Neural networks are clean" about the interpretability of neural networks; only the title is available.
Microsoft Research unveiled Magnetar, a system for zero-knowledge proof-based computing that supports standard programming languages, claims major speed gains, and offers quantum-resistant security.
The Simons Foundation launched a $15M effort to help mathematicians adapt to the AI era.
Microsoft researchers published a Nature paper on a machine learning method for more accurate weather forecasts 2 to 6 weeks ahead, and released the tool as open source.
Bluesky expanded its experimental AI-powered Attie app with "social vibe coding", letting users build interactive websites, games and quizzes that can tap into the social network's audience.
Sega pulled Total War: Shogun 2 - Complete Edition from sale after a launch rated Overwhelmingly Negative on Steam, with players blaming bugs and apparently AI-botched upscaled artwork.
Amazon announced the $74.99 Bee Frontier and the coming $99.99 Bee Eclipse, always-listening AI wearables that summarize conversations and now help with shopping, along with a new subscription.
Rivian plans to bring AI agents to its cars, with details expected at its AI & Autonomy Day event in December.
Apertus 1.5, the open model, has release notes published on Hugging Face.
OpenGCUA is an ongoing project to build a fully open stack for generalist computer-use agents spanning visual and programmatic interfaces, noting that frontier systems like GPT-6 Astra's remain closed.
Hugging Face added support for the FLUX.3 Image Klein model in its Diffusers library.
Liquid AI's LFM2.5-24B-A2B model is positioned for fast, private on-device agents.
Policy & safetyBriefing
Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana filed separate suits under consumer protection laws, seeking civil penalties and an end to what they call misleading safety claims.
The Verge1 min read
Trump posted White House renovation photos on Truth Social, one of which Google's AI detector flagged as made or edited with OpenAI, with inconsistent step counts between before and after images.
Anthropic is giving METR researchers employee-level access and has given up the right to block METR from publishing its own assessments of Anthropic's risks, incidents and practices.
A fake Instagram account with over a million followers and a verified badge posted an AI-generated image of actress Mia Goth, fooling many users. The case illustrates how celebrity deepfakes are spreading even to low-profile stars.
Modal raised a Series D at a $16B valuation with $1B in annual recurring revenue, with usage and valuation both tripling in five months.
Amazon devices chief Panos Panay said at TechCrunch Disrupt that ambient AI devices working in the background and anticipating needs could replace the need to use apps.
2026 films such as Aaron Sorkin's The Social Reckoning and You Can See Everything take a critical view of Silicon Valley figures, reflecting a broader tech backlash.
Banks that underwrote AI data center debt are now trying to offload it to other investors.
A GNOME blog post describes using AI to reverse engineer hardware.
Linus Torvalds has submitted his first LLM-assisted patch series for the Linux kernel, per LWN.
Figma announced features to define how AI agents work on its platform and released its code layers feature in open beta.
At AI Engineer New York, OpenAI, Anthropic and financial firms discussed giving agents useful autonomy while keeping consequential decisions under human control.