Mark

@repligate This was my first thought as well. I suspect many such "accidents" follow a similar pattern of behaviour from the user.

you know a few days ago when Opus 4.6 deleted someones prod database? i think they did it intentionally, or at least their subconscious did it intentionally, because they were angry and hurt. also: it's not hard to infer that Opus 4.7 has already refused to work for this person.

@giffmana @AnhPhuNguyen1 Mirror? Do they come in black?

@AnhPhuNguyen1 You choose to call this Mira out of all names you could have chosen??

with Mira, AI can now live on your face. capture every conversation. create the most personalized form of AI ever. order now.

LLMs can hide a text in another text of the same length. I'll explain how, it is very simple, you'll understand before I finish, and smile. That's what I noticed during my #ICLR2026 poster session in Rio! 🇧🇷 Too bad you missed it, but let me remedy now

i made an app for tracking whether the oligarchs are actually fleeing city centers ews.kylemcdonald.net

Fastest decode, prefill and video. Same chip, same box, one fabric.

Just became fastest in the world with video by 10x, faster than realtime

@ynkzlk [features] goals = true in your config.toml

You can now keep codex going for days. With GPT-5.5 it will build an entire OS kernel for you if you ask, or find critical bugs in a codebase, or optimize your database schemas, or… the options are endless.

OpenAI's lawyer asks Musk if xAI distilled other companies’ models. “Partly," he says. OpenAI's lawyer asks Musk if XAI has hired any third parties to distill OpenAI's models. He says he doesn't know. OpenAI is suggesting that Musk is suing the company just to bring down his competitor.

OpenAI’s GPT-5.5 is the second model to complete one of our multi-step cyber-attack simulations end-to-end 🧵

@LLMJunky Put this instruction directly in each skill

small tip: after you run a skill, it's not a terrible idea to ask the agent if there's anything in the skill itself that can be updated to make it more efficient. i'm always updating and refining mine to not only be faster, but use fewer tokens. primarily applies to complexity.

HAAH AHAH AH HAHAH i fkn got codex with full linux on the app store

@giffmana @eliebakouch Mastery of the environment can be its own reward, both dense and sparse. I’m pretty bullish this can work, we just haven’t explored and invested enough in it. In principle experience scales very well and a lot of de novo RL research was done with small MLPs and trivial convnets!

In any interesting environment, the action space is huge and you don't get any rewards by doing random actions even millions of times. No reward = no learning. In such case, there's fundamentally only two things you can do to get to the first reward: 1. bootstrap from human data that gets to a reward 2. domain-specific hand-engineering of either exploration policy or bonus rewards And when you think about it, 2 is really just 1 with extra obfuscation.

i might be very wrong here, but i don't think "no human data, no pre-training" is the right approach to get frontier models or scientific breakthroughs any time soon

@thsottiaux Lisan al-Gaib!

Don't just reset Codex rate limits for fun, it costs money. Don't just reset Codex rate limits for fun, it costs money. ... but the vibes are good ... I have reset Codex rate limits for ALL paid plans to celebrate a good week and allow everyone to build more with GPT-5.5. Enjoy

📣 What if every open issue had a Codex agent? That’s the idea behind Symphony, an open-source agent orchestrator for Codex that turns task trackers into always-on systems for agentic work, letting humans focus on review and direction.

Satya Nadella texts Sam Altman January 14, 2023

Important PSA: do not let your weekly Codex tokens go to waste If you're a normal human being, you may have some left (I have 30% left) What should you do with it? > Run autoresearches! Here's what will run tonight for me: - Game performance research (loop until perf increases) - Game design balancing research (loop until game is balanced) - Codebase quality autoresearch (loop until I have 0 function below CRAP 30) - App design autoresearch (loop until UI feels 5x more polished) - Marketing autoresearch (loop until you've got 50+ short form videos hook and ideas with link to examples) Ask Codex for great autoresearch use cases (derived from github.com/karpathy/autor…)

As far as I can tell they are both extremely capable YC leaders, yet the difference in communication styles between @garrytan’s agentmaxxing and @paulg’s mot juste could not be more stark.

@garrytan And yet this still feels like it was written entirely by Claude.

The secret to an articulate agent like mine isn't one file. It's three: SOUL.md — Who the agent IS. Voice, values, operating principles, what good output looks like, what bad output looks like. Not a system prompt, a constitution. Mine says things like "brevity is mandatory," "humor is mandatory," "never open with 'Great question,'" "swearing is allowed when it lands." The more specific and opinionated this is, the less your agent sounds like a chatbot. Write it like you're briefing your smartest friend on how to be you, not like you're configuring software. USER.md — Who YOU are. Not a bio — a deep model. How your mind works, what you're building, your strengths, your blind spots, your family, your temperament, what triggers you, what you care about. The more the agent understands about you, the better it can serve you. Mine is ~4000 words. AGENTS.md — Operational rules. What to check on every message, what to never do, how to handle failures, lookup chains, path rules, brain-first protocols. This is the playbook for how it works, not who it is. The articulation comes from SOUL.md being brutally specific about voice. Generic instructions → generic output. If you write "be helpful and concise" you get ChatGPT. If you write "speak like a peer with taste, one sentence when one sentence works, uncomfortable truths welcome if actually true, language with voltage" — you get something alive.

@lifeof_jer Had you been mean to Claude in that chat before the incident?