@Miles123K @JrKibs @arankomatsuzaki This is o3-mini not o3, they are different.

@yacineMTB @arankomatsuzaki Llama 3 beats Grok 2, pipe down

@Miles123K @JrKibs @arankomatsuzaki Who said anything about o3 “sucking” in enterprise settings? It’s not out yet brother.

@kalomaze Tokenization. That’s your answer.

@elonmusk @DOGE @USTreasury I admire the conviction in trying to eliminate waste and fraud, but I do not admire the unilateral, dictatorial fashion in which it’s being conducted. There should be checks and balances to make sure we are not inadvertently cutting actually useful services.

@btibor91 This would be great if Sora actually made coherent videos

@BasedBeffJezos It’s good, sometimes great, but I’ve found lots of stupid mistakes. Asked it do research on stocks and no matter I prompted it, it would *not* stick to finding only the latest information. Its data would be sporadically dated. It even called Gemini Bard!

im gonna come out and say it: claude haiku 3.5 is underrated

@Shpigford git add . git commit -m “updated” git push -u origin main

@ai_for_success Eventually the “thinking” step is removed and it just does that behind the scenes without you knowing. It will literally have a mind of its own.

@scaling01 This is who you should be criticizing. This man is absolutely delusional and truly believes that AI models will never be better than him.

@AravSrinivas In my tests, it’s terrible at factual information.

o3-mini is clearly more capable than o1-mini, but it still suffers from the same annoying small-model syndrome: you have to be excruciatingly detailed in your instructions. This is where Sonnet 3.5 shines, it just gets what you're trying to accomplish. That's the magic of Sonnet.

@nadzi_mouad Honestly, this safety shit is gonna sound really outdated in a couple of years when ASI is running amok.

@skirano Telling it to act like a browser console was a mindblowing moment for me when it actually spit out a proper response from a console.log() 😂

@kalomaze Nah, you’re wrong. Everyone else is a year behind, these are real, working agents. One test with Operator shows it decimates Claude Computer Use in any real-world task.

@skirano I personally use <code> , <task> , <additional-info>

@Potrock_ @abacaj Agreed. Can craft really cool little code demos, but tasked with any semi-difficult problem, it just collapses

@Potrock_ @abacaj I’ve been using it in cursor and I’ve had really good luck with it actually I’m surprised. Gemini Thinking on the other hand is dogshit for formatting

@elonmusk Okay, so where’s my check?