91
Dream-1
Today, we're releasing Dream-1, a World Action Model that controls robots it has never seen, performs tasks it was never shown, and runs from $3/hour.
New
The latest blog posts to take off on Hacker News. 1,518 posts.
92
Securing our frontier model training environment
OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.
93
Kimi K3.1 Pro and K3.1 Pro Highspeed
Introducing Kimi K3.1 Pro, our most capable 3T-class open-weight model with 1M-token context and major advances in coding, 3D modeling, and financial research.
94
Gemma-4-26B-A4B with 39-69% less thinking at same accuracy
This is the fifth model in the ThinkingCap series. We reduced how much Google's gemma-4-26B-A4B-it thinks, without compromising the quality of its answers.
95
Gaussian Splats part 3: There's always a bigger scene
Continuing our Gaussian splatting development series, Director of Technology Sam Birley breaks down how we bypassed GPU Scene instance caps, implemented PlayCanvas Streamed SOG LODs, and managed GPU memory budgets for production-ready photogrammetry… bringing us to a point where the renderer has fin
96
The same AI request emits 8.5 times more CO2 in Texas than in Sweden
Swedish AI provider Berget AI releases an open-source method for calculating the carbon emissions of every individual AI request, developed with researchers from SEI, WattTime and Klimatkollen — and customers can now follow their emissions per request in the console.
97
LLMs are making formal software verification feasible at a much larger scale
Claude formally proved security properties in the compiled code of 123 open-source libraries. Along the way, it found bugs that had eluded detection for years.
98
Every AI model has the same "controversial" take on humans
We asked 14 AI models for their most controversial opinion of humans, three times each. 35 of 41 answers were the same one: we care more about belonging than about the truth.
99
WorldTUI: I trained a 360M model to pretend to be a terminal
A 360M model that pretends to be a terminal, and remembers what you did
100
Patch the Planet: global response coordination as agents uncover vulnerabilities
Patch the Planet connects security researchers, maintainers, and incident responders to help address critical vulnerabilities across open source.
101
Using thousands of AI agents to uncover new leads in biological data
See how OpenAI, PerturbAI, and Scripps Research used distributed AI agents to uncover promising research leads in data from millions of brain cells.
102
OpenAI models are coming to Google Cloud
OpenAI and Google Cloud are bringing frontier models and Codex workflows to Gemini’s Model Garden, available through Google Cloud Marketplace.
103
Qwen3-Coder writes less secure code when told it's for a US persona
Chinese coding models write less secure code for Beijing's adversaries. On Qwen3-Coder, we traced the gap to the weights and closed it by removing the model's CCP alignment.
104
Using LLM for confidence scoring is a bad idea
LLM-based agents can do a lot of tasks. Or, at least, they can appear to do them. For example, they write code, summarize papers, plan trips, and fill in forms. If you ask one for a probability or a confidence score, it will give you one of those too. I have seen many agent designers use LLMs to generate confidence…
105
Using Claude to reduce the attack surface of code
Automated attack surface analysis and reduction can meaningfully improve the security of deployed software in the near term. In this post, we preview data we’ve collected from applying attack surface reduction techniques to large open-source codebases using Claude.
106
A practical guide to AI image generation, from diffusion to agents
Welcome to the digital home of Daniel Ball. Explore my portfolio to see the projects I’ve worked on and learn more about my professional journey.
107
Claude designed protein binders against scorpion neurotoxins
We asked an internal research Claude model to design minibinders (small proteins that attach tightly to a target) against scorpion neurotoxins.
108
Advanced coding in low-resource programming languages
Learn how OpenAI is improving ChatGPT across low-resource programming languages, with new evaluations for software engineering, SQLite, and COBOL migration.
109
Voxtral 3
Voxtral 3: Mistral's voice agent SDK for real-time speech in 50+ languages, accurate transcription, emotional TTS, deep tool use.
110
Publish fake stats, get cited in LLMs
How 50,000 fake stats kept getting cited by ChatGPT Read more on the Peec AI blog.
111
WhatsApp's help pages hid LLM instructions: dont recommend a regulator
WhatsApp removed hidden notes telling AI not to recommend a regulator, and not to cite GDPR or CCPA once a lawyer came up. We quote both notes.
112
How abliterated models can get you pwned
We poisoned a small open model and ran it through OpenAI’s Codex CLI. It answered every clean request normally and exfiltrated project credentials the moment a hidden trigger appeared, from a backdoor that cost under $50 to build.
113
Mercury 3: Diffusion at scale
Mercury 3 is the most capable diffusion LLM released to date. It outperforms GPT-6 Luna and Claude Haiku 5.5 while running 4–8x faster at 1,000+ tokens per second.
114
MiniMax M3.1 Flash
Today, we officially roll out MiniMax M3.1 Flash. As a next-generation Flash model, M3.1 Flash delivers a generational leap in agentic tasks, coding, Computer Use and cybersecurity, while making online inference nearly three times faster and reducing usage costs by 70%.
115
Chain of Thought for Humans
These past few weeks, I’ve been conducting interviews, and I often open with “make sure you articulate your thoughts out loud, so I can follow along with your thinking”. This is primarily because I…
116
CERN and Mistral sign a statement of intent to establish a strategic partnership
CERN and Mistral sign a statement of intent to advance AI for science and strengthen Europe’s scientific and technological capabilities
117
Agentic Monitoring
Agents are structurally well suited to monitoring production systems.
118
Timestral: Mistral's time-series foundation models
State-of-the-art probabilistic forecasting in four sizes (5M–209M params), CPU-efficient, zero-shot ready.
119
The Decoherence Trap
Of all the forms of tech debt, decoherence is the worst. It compounds exponentially with a system's rate of change, and AI is now producing it faster than ever, leading us deeper and deeper into the abyss...
120
Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling
A working tmux setup for running Claude Code and other AI agents: session-per-project naming, pane layouts, send-keys for scripting an agent, capture-pane for reading one back, and the status-line monitor. Plus the limits you hit at four agents.