Jay-F. 😎

If Claude Mythos doesn’t meet the hype, Anthropic will lose market share.

Anthropic pays $750,000+ a year for engineers who can build LLM architectures from scratch. Stanford taught the entire thing in 1 hour lecture & released it for free. Bookmark & watch this today before someone takes it down and read this article below

@BredsguardDalen Grandpa must have been a point near medically insane

Beach hacks to start off the summer

@elonmusk @jamesdouma Violence man. Unnecessary violence.

@jamesdouma What are the most important things we need to fix?

Been making stuff with grok build this last week. The world is feeling very post-scarcity right now.

@tomfgoodwin what agent are you mostly using?

errmmmmm, not to be miserable but has anyone noticed that agentic AI doesn't really work at all. Like the errors compound, fragile integrations ( any external change breaks it ) , observability is an issue, no verification, context loss, the whole thing seems VERY tricky Not sure this can ever be fixed.

The downside of frontier LLMs is that they’re engineered to be too polite.. They lack the human capacity for rudeness. I need a model with Russian energy. That raw truth. Have you seen an ai model insults? Even a five year old has better banters.

Grok voice command recognition could do with a lot more work. Feels slow in the head.

Down to 36% hourly already.

The Codex usage limits have been reset for all paid ChatGPT subscriptions. You should be back to 100% weekly and 100% hourly limits. Let the tokens do incredible things today and have fun.

Grok is proud!!! Always feels it’s better than other models. smh🤦🏻‍♂️

Womanize.

Comment if you want the one line command

Codex is better when it remembers how you work, has a very detailed high quality structure for solving tasks and actually uses the skills you have. so i open sourced my skills pack. 120 skills. one command install. works on mac, linux, windows. not prompts. actual reusable workflows. You’ll see the difference immediately.

As a male If a guy you don’t know compliments you, first thought is “he’s gay” second thought is “might be gay” third thought is “maybe he’s bi”

Literally gave out the keys to his kingdom

Still the best ancient game.

Is codex your best bet or you enjoy playing all sides?

I don’t know about 24+ hours but I can speak on running tasks for hours or through midnight to morning. Set /goal. Give me tasks that lead to goal - in other words a lot of context. Give it elevated permissions or full access. (This makes sure it doesn’t pause to start asking your for permissions). Note that when giving an agent full access, it’s best to always monitor it but using /goal in the same context essentially gives it focus and makes it more reliable.

People that run 24+ hours Codex tasks Can you share what you’re running exactly? Everyone is sharing the hours but not the task itself, I feel that most of them are just engagement baits

Yesterday I said GPT-5.5 edges out Opus 4.8 for coding. The thread blew up, most folks agreed, but some pushed back hard. Then DeepSWE (a hard long-horizon coding benchmark) dropped its numbers: → GPT-5.5: 70% pass@1, #1 → Opus 4.8: 58% And GPT-5.5 gets there with ~2x faster runs, ~½ the cost, and ~⅓ the output tokens. One benchmark isn't everything. But smarter, cheaper, and faster all at once? Hard to argue with.