Aleksey Smolenchuk

Prism is now powered by Codex, giving it real agentic superpowers. Try it out! prism.openai.com

Once in a while I play with my Apple IIe just to feel the contrast between how we code now and how we used to code

This is actually insane. Dude hard-coded a WebAssembly (WASM) interpreter into the weights of a transformer, losslessly. In essence, a computer is running inside a LLM that can actually run computations, not infer or guess a calculation like most do today.

Working at OpenAI is fun because questioning everything and taking risks is part of the culture. Within Codex, the team asks itself how we could make it an order of magnitude better every few months and then sets most things aside to go and do it across the entire stack. Some examples were the Codex App and our first deployment of Cerebras inference with WebSockets. We are now well under way on the next bet and it’s making even our best engineers nervous as it’s at the edge of what’s possible today.

a small window into the opportunity of AGI

@steipete challenge accepted

NOT ME I SWEAR

🔥Two things I'm esp excited about 5.4: 1. Unification: we merged our codex & mainline models 2. Efficiency: we brought the efficiency of 5.3-codex to CUA & knowledge work. We only showed 3 such plots in the blog but many of our evals required less time (tokens/tools) than 5.2. What should we fix for the next model?

gpt-5.4 is better than codex-5.3 at following and retaining identity and personality related instructions, so you can have it drive both openclaw’s main thread and the builder/coder sub-agents, and you get both incredible coding performance and actual personality.

GPT-5.4 is launching, available now in the API and Codex and rolling out over the course of the day in ChatGPT. It's much better at knowledge work and web search, and it has native computer use capabilities. You can steer it mid-response, and it supports 1m tokens of context.

@nodp53 That shouldn’t happen. Let’s see if we can get it fixed quickly.

I would really love to use this tool! However, my biggest issue is that it still loses all in-text citations after a single PDF recompilation (as of March 5, 2026). Steps to reproduce: I upload my structured LaTeX paper (IMRD format, one .tex file per section, figures in a dedicated directory, and references in a .bib file). This exact setup compiles flawlessly on both Linux and macOS. But in Prism, after recompiling the PDF, every single citation marker still turns into '???' signs.

Prism is now powered by Codex, giving it real agentic superpowers. Try it out! prism.openai.com

@lxe Studying math with this tool is incredible. I upload the lecture transcript, slides, and notes, ask it to explain everything as if I'm a beginner, with examples for each concept, and it generates a complete PDF for me to read on my Kindle. We are truly living in the future.

We just posted a new preprint: “Single-minus graviton tree amplitudes are nonzero.” Yes: a helicity sector long assumed to vanish in quantum gravity can actually appear under well-defined kinematics. Preprint: cdn.openai.com/graviton/gravi…

🧵1/ We've brought the most advanced AI to Prism by introducing Codex to Prism. Prism is already the best place for scientific writing to happen—and with Codex, now you can write, compute, analyze, and iterate all in one place.

We integrated the Codex harness into Prism (prism.openai.com) — this means you get skills, reasoning levels, and the raw tenacity of the Codex model in your LaTeX environment. Oh and we also built version mgmt into Prism, which was one of the top requests. See below thread from @vicapow for some great examples of the power of Codex inside Prism 👇

My new hobby is getting these *huge* bare metal servers for $59 on Hetzner:

netbooks are back

@rauchg At first glance it seems that this failure mode has very little to do with AI and a lot with ID collisions… …but then on second thought if even the mode collision-resistant id appears in the training data, there’s a non-random chance it could get generated

A Vercel user reported an issue that sounded extremely scary. An unknown GitHub OSS codebase being deployed to their team. We, of course, took the report extremely seriously and began an investigation. Security and infra engineering engaged. Turns out Opus 4.6 *hallucinated a public repository ID* and used our API to deploy it. Luckily for this user, the repository was harmless and random. The JSON payload looked like this: "𝚐𝚒𝚝𝚂𝚘𝚞𝚛𝚌𝚎": { "𝚝𝚢𝚙𝚎": "𝚐𝚒𝚝𝚑𝚞𝚋", "𝚛𝚎𝚙𝚘𝙸𝚍": "𝟿𝟷𝟹𝟿𝟹𝟿𝟺𝟶𝟷", // ⚠️ 𝚑𝚊𝚕𝚕𝚞𝚌𝚒𝚗𝚊𝚝𝚎𝚍 "𝚛𝚎𝚏": "𝚖𝚊𝚒𝚗" } When the user asked the agent to explain the failure, it confessed: The agent never looked up the GitHub repo ID via the GitHub API. There are zero GitHub API calls in the session before the first rogue deployment. The number 913939401 appears for the first time at line 877 — the agent fabricated it entirely. The agent knew the correct project ID (prj_▒▒▒▒▒▒) and project name (▒▒▒▒▒▒) but invented a plausible-looking numeric repo ID rather than looking it up. Some takeaways: ▪️ Even the smartest models have bizarre failure modes that are very different from ours. Humans make lots of mistakes, but certainly not make up a random repo id. ▪️ Powerful APIs create additional risks for agents. The API exist to import and deploy legitimate code, but not if the agent decides to hallucinate what code to deploy! ▪️ Thus, it's likely the agent would have had better results had it not decided to use the API and stuck with CLI or MCP. This reinforces our commitment to make Vercel the most secure platform for agentic engineering. Through deeper integrations with tools like Claude Code and additional guardrails, we're confident security and privacy will be upheld. Note: the repo id above is randomized for privacy reasons.

bird

@VadimStrizheus @openclaw you’re doing it wrong

Top 10 best things on earth.