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.
How This AI Breakthrough with Pure Mathematics and Reinforcement Learning Could Help Predict Future Crises
An artificial intelligence breakthrough uses reinforcement learning to tackle the Andrews-Curtis conjecture, solving long-standing counterexamples and hinting at tools for forecasting stock crashes, diseases and climate disasters
Deep learning-based prediction of the selection factors for quantifying selection in immune receptor repertoires
Nature Machine Intelligence – TCRsep, a deep learning model for predicting selection factors that quantifies the T cell selection process, is introduced. Also, various benchmarks are designed to…
Kolmogorov–Arnold graph neural networks for molecular property prediction
Nature Machine Intelligence – Li et al. developed KA-GNNs, graph neural network architectures enhanced by Kolmogorov–Arnold networks, which improve accuracy and interpretability in molecular…
NASA and Google Collaborate on AI Doctor for Mars Trip
The tool, which uses Google Cloud’s Vertex AI environment and a mixture of open-source large language models (LLMs), is specifically tailored to the medical challenges of long-distance space travel.
AI Safety at the Frontier: Paper Highlights, July ’25
tl;dr Paper of the month: • Language models can transmit behavioral traits like misalignment through seemingly unrelated data (pure number sequences)…
OpenAI’s AWS partnership goes deeper with open models designed for AI agents, coding, and long-form thinking
Optimized for low-resource environments, but benchmarks remain unclear
Computer Science Grads Struggle to Find Jobs in the A.I. Age
As companies like Amazon and Microsoft lay off workers and embrace A.I. coding tools, computer science graduates say they’re struggling to land tech jobs.
Trump’s AI Strategy Is At War With Itself
Suresh Venkatasubramanian, Costas Samaras, and Cole Donovan say the stated goals of the AI Action Plan conflict with the Trump administration’s actions.
One Year On, EU AI Act Collides with New Political Reality
Caterina Rodelli and Sarah Chander revisit the EU AI Act as deregulation and militarization threaten its promise to protect rights and regulate harmful AI use.
The US Aims to Win the AI Race, But China Wants to Win Friends First
Charles Mok contrasts the “America First” Trump AI Action Plan with China’s plan to lead on global AI governance.
Blueberry Hill
ChatGPT 5 was released today. ChatGPT-maker OpenAI has unveiled the long-awaited latest version of its artificial intelligence (AI) chatbot, GPT-5, saying it can provide PhD-level expertise. Billed as “smarter, faster,...
My Lethal Trifecta talk at the Bay Area AI Security Meetup
I gave a talk on Wednesday at the Bay Area AI Security Meetup about prompt injection, the lethal trifecta and the challenges of securing systems that use MCP. It wasn’t...
The Reality of Managing 10 AI Agents in Production: What We’ve Learned Building Our AI-First Revenue Team at SaaStr
By the end of 2025, we’ll have 10 distinct AI agents running in production at SaaStr. Maybe more. Not as a tech experiment or marketing stunt, but as core members...
From bootcamp to bust: How AI is upending the software development industry
Coding bootcamps have been a mainstay in Silicon Valley for more than a decade. Now, as AI eliminates the kind of entry-level roles for which they trained people, they’re disappearing.
We can still sleep peacefully — or so we thought. Steganography via internal activations is already possible in small language models — a potential first step toward persistent hidden reasoning.
Introduction In this post we continue to work on the project, started in our previous post, of exploring hidden reasoning of large language models…
Apple brings OpenAI’s GPT-5 to iOS and macOS
It’s unclear exactly how GPT-5’s new approach to model-switching will work here.
GPT-5 Under Fire: Red Teaming OpenAI’s Latest Model Reveals Surprising Weaknesses
How secure is GPT-5 out of the box? Here’s how it fared against real-world threats and why alignment must be earned.
The Three Computer Solution: Powering the Next Wave of AI Robotics
Physical AI, embodied by industrial systems such as humanoids and factories, is being accelerated by three NVIDIA computers and software platforms across training, simulation and inference.
Tensions rise as Nvidia races to calm China over controversial AI chips built for surveillance
H20 backlash leads to questions over Nvidia’s future in China
AI industry horrified to face largest copyright class action ever certified
Copyright class actions could financially ruin AI industry, trade groups say.
OpenAI’s GPT-5 rollout is not going smoothly
It also failed on a simple algebra arithmetic problem that elementary schoolers could probably nail, 5.9 = x + 5.11.
From Knowledge to Action by @ttunguz
GPT-5 launched yesterday. 94.6% on AIME 2025. 74.9% on SWE-bench. As we approach the upper bounds of these benchmarks, they die. What makes GPT-5 and the next generation of models...
Skip The Hype Reel — GPT-5’s Real Story Is In The System Card
Chat GPT-5 doesn’t have crowd-pleasing upgrades, but they do directly address operational pain points enterprises face today with GenAI apps.
Introducing Open SWE: An Open-Source Asynchronous Coding Agent
The use of AI in software engineering has evolved over the past two years. It started as autocomplete, then went to a copilot in an IDE, and in the fast...
Introducing AI Sheets: a tool to work with datasets using open AI models!
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
How enterprises can transition their knowledge and systems for Agentic AI
Five key steps to future-proof an enterprise for the agentic AI era
How Google’s Genie 3 could change AI video – and let you build your own interactive worlds
AI is becoming increasingly adept at generating virtual environments that simulate real-world physics and respond dynamically to users.
Forklift certified: Bedrock secures $80M to develop AI equipment operators
Bedrock Robotics is developing a new generation of autonomous heavy equipment equipment operators who, they hope, will be able to work 24/7.
The GPT-5 Launch Was Concerning
There were screenshots of a classic LLM issue floating around Bluesky after the GPT-5 launch yesterday, and I asked GPT-5 myself to confirm.
CoT May Be Highly Informative Despite “Unfaithfulness”
Recent work from Anthropic and others claims that LLMs’ chains of thoughts can be “unfaithful”. These papers make an important point: you can’t take everything in the CoT at face...
GPT-5 vs GPT-4: Here’s what’s different (and what’s not) in ChatGPT’s latest upgrade
The new king is here, so how does it compare?
SoftBank buys Foxconn’s Ohio plant to advance Stargate AI push, Bloomberg News reports
SoftBank Group Corp is acquiring Foxconn Technology Group’s electric vehicle plant in Ohio, in a bid to launch the Japanese company’s Stargate data center project, Bloomberg News reported on Friday.
Google is spending $1 billion on boosting AI training at US universities
Google wants to help train students in AI
The New ChatGPT Resets the AI Race
With GPT-5, OpenAI is making its strongest effort yet to hook users.
GPT-5 Is Launched But Set Aside Your Expectations That This Was Going To Be Either AGI Or Artificial Superintelligence Since It Clearly Isn’t
I take a look at GPT-5, newly released today, and provide some initial comments. It is a handy improvement. But it surely isn’t AGI or ASI. Here’s the AI Insider...
OpenAI’s most powerful AI model is here and free for everyone
OpenAI has officially launched its highly-anticipated GPT-5 model. The model is faster, more accurate, and hallucinates less.
OpenAI’s new open-source model is basically Phi-5
OpenAI just released its first ever open-source large language models, called gpt-oss-120b and gpt-oss-20b. You can talk to them here. Are they good models…
Forget London or Berlin—OpenAI builds one of Europe’s fastest AI hubs in a quiet Nordic town
Europe’s most powerful data center rises in the Arctic Circle
OpenAI Releases GPT-5
We are introducing GPT‑5, our best AI system yet. GPT‑5 is a significant leap in intelligence over all our previous models, featuring state-of-the-ar…
OpenAI releases GPT-5, a potential barometer for whether artificial intelligence hype is justified
OpenAI has released the fifth generation of the AI technology behind ChatGPT. The update is being closely watched to see if generative AI is advancing or plateauing.