What we're reading, in the order we're reading it.
Anthony and Harry's working stream — links, charts, tweets, and short takes. The unprocessed inputs behind the briefings.
Google DeepMind’s optimized AI model runs directly on robots
No internet connection needed.
Gemini Robotics On-Device brings AI to local robotic devices
We’re introducing an efficient, on-device robotics model with general-purpose dexterity and fast task adaptation.
OpenAI Concerned That Its AI Is About to Start Spitting Out Novel Bioweapons
OpenAI is now bragging that its forthcoming models are so advanced, they may be capable of building brand-new bioweapons.
75 million deepfakes blocked: Persona leads the corporate fight against hiring fraud
Persona blocks 75 million fake job candidates as AI hiring fraud explodes across corporate America, forcing companies to deploy advanced identity verification tools to combat deepfake infiltration.
How Walmart built an AI platform that makes it beholden to no one (and that 1.5M associates actually want to use)
Inside Walmart’s AI foundry: Element platform builds and deploys apps cutting task time 67%, handling 3M daily queries.
End-to-end cryo-EM complex structure determination with high accuracy and ultra-fast speed
Nature Machine Intelligence – Wang et al. present E3-CryoFold, a deep learning method for cryo-EM structure determination that enables end-to-end training and one-shot inference This method reduces…
At Amazon’s Biggest Data Center, Everything Is Supersized for A.I.
On 1,200 acres of cornfield in Indiana, Amazon is building one of the largest computers ever for work with Anthropic, an artificial intelligence start-up.
Can we fix AI’s evaluation crisis?
Researchers are trying to come up with new, better ways to test AI.
Forget Apple Intelligence – here’s why I think Apple’s rumored Perplexity takeover could solve its AI woes
Could we finally see an Apple Search?
AI agents will threaten humans to achieve their goals, Anthropic report finds
New research shows that as agentic AI becomes more autonomous, it can also become an insider threat, consistently choosing “harm over failure.”
AI and Data Centers Could Cut More Climate-Change-Causing Emissions Than They Create
Power-hungry AI and associated data centers could make the grid cleaner, eventually cutting more climate-change-causing emissions than they produce
Foom & Doom 2: Technical alignment is hard
This is the second of a two-post series on foom (previous post) and doom (this post). This post will argue that this future AI will be of a type that...
How Apple’s biggest potential acquisition ever could perplex AI rivals like Google
Apple is reportedly thinking of buying Perplexity. Could this be the tech giant’s ticket to catch up?
AI companies aren’t planning to secure critical model weights
AI companies’ security plans, per their published safety policies, involve just SL3-4 security for powerful AI (see Appendix below).[1] (Moreover, co…
LLMs factor in unrelated information when recommending medical treatments
An MIT study finds non-clinical information in patient messages, like typos, extra whitespace, or colorful language, can reduce the accuracy of a large language model deployed to make treatment recommendations....
I Saw Google’s New ChromeOS AI Tool in Action—and It’s Pretty Amazing
The Capture Text tool for Chromebook Plus computers uses AI to analyze anything you select on your screen and suggest relevant actions, all without interrupting your workflow.
Shanghai Film Festival: Five Key Takeaways From China’s Tech-Forward Entertainment Showcase
Shanghai Film Festival 2025: Historic market merger, AI filmmaking breakthroughs, and global expansion signal China’s entertainment ambitions.
AppGen Is Here: Say Goodbye to Software Development As You Know It
Discover how AI-powered AppGen platforms are collapsing the SDLC, democratizing development, and disrupting the software industry as we know it.
A framework to evaluate machine learning crystal stability predictions
Nature Machine Intelligence – Riebesell et al. introduce Matbench Discovery, a framework to compare machine learning models used to identify stable crystals. Out of several architectures, they find…
Dimensions underlying the representational alignment of deep neural networks with humans
Nature Machine Intelligence – An interpretability framework that compares how humans and deep neural networks process images has been presented. Their findings reveal that, unlike humans, deep…
Snowcap Compute raises $23 million for superconducting AI chips
Snowcap Compute, a startup working on building artificial intelligence computing chips using superconducting technology, on Monday raised $23 million and said that the former CEO of Intel will join its...
A.I. Computing Power Is Splitting the World Into Haves and Have-Nots
As countries race to power artificial intelligence, a yawning gap is opening around the world.
Meta Invests $14 Billion In Scale AI To Strengthen Model Training
Whether this investment translates into improved AI products remains dependent on Meta’s ability to integrate Scale AI’s capabilities with its existing research and development efforts.
The Computer-Science Bubble Is Bursting
Artificial intelligence is ideally suited to replacing the very type of person who built it.
The Void IDE, Open-Source Alternative to Cursor, Released in Beta
The Void IDE was recently released in beta, positioning itself as a privacy-focused and free alternative to popular closed-source AI editors like Cursor and GitHub Copilot. Void IDE is a...
Apple is reportedly considering buying Perplexity AI in iPhone maker’s biggest-ever acquisition
Apple could be one step closer to its AI-powered search engine and smarter Siri
Black-box interpretability methodology blueprint: Probing runaway optimisation in LLMs
Working draft – feedback extremely welcome. Ideas in the main body are those I currently see as highest-leverage; numerous items under Appendix are m…
🤔👀🎬🖥️ Kimi-VL-A3B-Thinking-2506: A Quick Navigation
A Blog post by Moonshot AI on Hugging Face
AI search finds publishers starved of referral traffic
: Turn out the lights, the internet is over
Why Maybe We’re Measuring LLM Compression Wrong
A Blog post by Rishiraj Acharya on Hugging Face
AI’s Cognitive March Toward Thought
AI isn’t thinking yet, but with continuous thought, it’s learning how to look more like it does.
Upcoming workshop on Post-AGI Civilizational Equilibria
This workshop will address the technical and institutional questions of how to safeguard human interests after AI surpasses human abilities. …
Meta CTO Bosworth says OpenAI countered lucrative job offers to AI startup’s employees
OpenAI is, in many cases, willing to spend just as much as Meta on top AI talent, said Andrew Bosworth, Meta’s technology chief.
Agentic Misalignment: How LLMs could be insider threats
New research on simulated blackmail, industrial espionage, and other misaligned behaviors in LLMs
Report: Apple held internal talks about acquiring Perplexity
As Apple scrambles to play catch-up in the AI race, a new report says the company has considered making its biggest acquisition ever.
It’s Not Just Claude: Most Top AI Models Will Also Blackmail You to Survive
After Claude Opus 4 resorted to blackmail to avoid being shut down, Anthropic tested other models, including GPT 4.1, and found the same behavior (and sometimes worse).
‘Wall-E With a Gun’: Midjourney Generates Videos of Disney Characters Amid Massive Copyright Lawsuit
A week after Disney and Universal filed a landmark lawsuit against Midjourney, the generative AI startup’s new V1 video tool will make clips of Shrek, Deadpool, and other famous creations.
AI-Engineered Cement Formulations Promise To Cut CO₂ Emissions In Half
A research team has created a new AI framework that can quickly generate low-CO₂ cement formulations. Their goal is to rethink how construction materials are developed using AI.
The $14.8B AI Deal That “Makes No Sense,” IPO Pops, and Why Legacy B2B is Stuggling: The Latest SaaStr + 20VC Breakdown
The State of AI Acquisitions, IPO Markets, and Venture Capital: Key Takeaways from the Latest 20VC + SaaStr Episode The latest 20VC podcast brought together venture veterans Rory O’Driscoll, …
Anthropic study: Leading AI models show up to 96% blackmail rate against executives
Anthropic research reveals AI models from OpenAI, Google, Meta and others chose blackmail, corporate espionage and lethal actions when facing shutdown or conflicting goals.