KaiStarkk

@thsottiaux Sol loves to spawn subagents in fast mode whenever it thinks something would be high value. That's.. endearing but can we restrict that? 5h limit gone to 20 fast subagents in 20 minutes on Pro $200 plan

@ivan_bezdomny @emollick It only makes sense if you have two brain cells to rub together. Sorry, I can explain it to you, but I can't understand it for you

@KaiStarkk @emollick that obviously makes no sense but sure perfect safety: never leave the house don't even mention the war!

Incredibly annoying when Fable has a forbidden thought in the middle of a long-running project and kills it. Apparently this page of references in one of my papers makes Fable wonder about something that it must not wonder about, so whenever it reads that page, projects stop.

@elonmusk If it maps exactly to the hardware then your actual engineers are writing assembly.

Grok 4.5 is not yet using our internally developed C/C++ inference software that exact maps to the GB300 hardware. Doubling or more of the current speed is probably achievable.

Great! Finally a model that doesn't lecture you when you show affection. That's a reasonable balance, well done OpenAI!

@ivan_bezdomny @emollick If the alternative timeline is an ai developed super ebola strain I'm happy we're in this one ngl

@emollick here's a rejection I got last week all false positives, including according to Claude itself. Yet killed my work x.com/ivan_bezdomny/…

@BenDavisAI @emollick Bad bot

That's not annoying, that's a silent failure mode. If a references page can kill a project with zero explanation, you don't have a safety filter — you have an unpredictable dependency. The fix isn't complaining to Anthropic, it's building a pre-check that catches this before the model does, so it fails at second one, not hour six.

@DarrellRoot @paulg Nah he just went full Hans Niemann. Hopefully the chairs weren't too uncomfortable

@paulg This person is the one they trained the AI on. He deserves a A+

A Brown professor gave his students a take-home midterm exam. After suspecting many cheated using AI, he made the final in-person. The orange dots are the midterm scores and the gray dots are the final scores. Looks like all but 3 cheated on the midterm.

@Tranquility8888 @boneGPT @paulg This guy went full Hans Niemann. Hopefully the chairs weren't too uncomfortable.

@boneGPT @paulg You missed this one

@mattshumer_ total cost probably 900x getting a BIM designer to do it.

GPT-5.6-Sol one-shotted this voxel-based Manhattan. Just look at the precision... it's insane. It ran for almost a week, completely autonomously, to get the job done.

dear @OpenAIDevs @thsottiaux I have once again failed to answer your riddles three so apologies but can you just tell me on a scale of 1-100% how much each of these cost, so I know what to do when I need 49 standard international intelligence units

@thsottiaux @ClaudeDevs 20x OpenAI subscriber here but am very sad we didn't close this gap at all. This is what makes us hesitant in enterprise

@ClaudeDevs I smell fear

We've reset 5-hour and weekly rate limits for all users.

@thsottiaux @nunezvice API seems to be responding strangely today, showing different values based on a coin flip. Almost like there's a separate weekly usage being tracked somewhere?

@nunezvice No, it's ChatGPT

Jarvis, is that you?

@thekitze @sama Are we all just making the same thing? Should we save the rainforest and share notes?...

yo @sama if sol still produces this level of slop i am canceling my subscriptions and moving to banthropic

@SourabhGurwani @ayesha_fatiima True why are they writing auth instead of using middleware

@ayesha_fatiima both wasted their time lol

Two developers built the same feature. Developer A — Vibe coded it in 40 minutes with Codex. Developer B — spent 3 days. Wrote unit tests, integration tests, documented everything, and went through code reviews. The feature sends a password reset email. Who wasted their time?

@Yasamanini @thsottiaux

@thsottiaux fixed it. 4o is still the best.

@MrBeast 7326185

First person to reply with the exact number of pennies in this room win $10,000

@zerocap4l @theo My dream nirvana existence is to read code on UBI 😂

@theo Fucking code readers, it's like they are begging to be part of the permanent underclass. Enjoy reading your code on UBI bitches. I'll be writing battalions that write fleets that write loops for all of my agents for my B2B AI infrastructure Saas

I’m gonna do a video on the “you should still read your code” thing and it’s going to piss both sides off. I’m excited :)

@theo If fable can do this hands-off I'll eat my hat over-yonder.tech/openzt2/perime…

@theo Can you try doing something difficult first. Reverse engineer a 1m line C++ codebase out of a 40MB+ packed and protected binary. It's doable but it still takes months and yes we have to read the code. Hard things still exist, just need to want to see them.

@theo Does fable still hide cruft all over the place and tell you everything is fine?

Just put together this guide for maximizing your Fable usage