Stepan Goncharov

Everything is possible on @X

@NeuraPod Are you saying it's not only for robots???

Having a Neuralink connected to an Optimus prosthetic will be the next step in increasing freedom for paralyzed individuals. 🦾

@buddyhadry Name is 🔥

hear me out. elonMUX, the desktop app for grok build

@dimabytes VCS 💪

@stepango We have shared repo with skills, and devs can do review there and improve it. Not pair programming, but close hahah. How do you share skills in your team?

Are you doing pair prompt writing already?

@eboziev Pleasure is all mine🫡

@stepango Thanks man 🤝😎

Welcome to the club💰

Oh, I see where this is going 😏

@JonasBadalic So sweet 💗

@stepango Yes, with my wife

@thesoragirls @morganlinton @VulcanBench If not refuses your prompt 🤣

@stepango @morganlinton @VulcanBench that's insane. so Fable is way cheaper??

I benchmarked Fable 5 Low effort vs GLM 5.2 High effort and Sonnet 5 High effort using @VulcanBench 🖖 Evals were on 10 real engineering tasks from flask, aiohttp, sqlglot + more. All post-training-cutoff, so nothing memorized. All three scored 80%. The costs were $2.27 vs $8.41 vs $15.81. And the wildest thing is which model was the most expensive, and it wasn't Fable 👀 Results in first comment below 🔽

🤔

Video production quality of @SpaceX is unmatched 🤩

@naoufal_elh @cursor_ai Welcome 🚀

Excited to share a personal update, I have joined @cursor_ai as a Field Engineer.

@JonDevX This number of tokens, if used correctly, can probably build entire Metaverse and make it into a good product 🤣

@stepango What the hell were rhey burning so many tokens on??? That's mad amounts, internal tooling can't be THAT expensive... Likely they were just inefficient

Now layoffs make sense, right?

@LGrapenthin @kenshii_ai Forwarded to the team 🙏

@stepango @kenshii_ai grok.com/share/bGVnYWN5…

Grok is becoming the most frustrating AI on the market. I used to love using it, but Grok 4.3 feels like a completely different product. It gets trapped in obvious logical fallacies, doubles down when challenged, and poisons the rest of the conversation by treating bad assumptions as facts. At the same time, the growing limitations and restrictions make the experience even worse. It feels like you're constantly running into walls instead of having a real conversation. Mistakes aren't the problem. Every model makes them. The problem is when a model refuses to admit it's wrong. An AI that can't course-correct stops being useful very quickly.

@LGrapenthin @kenshii_ai And still, concrete examples would've been much more helpful🙏

@stepango @kenshii_ai Admit it you are just running a quantized version 4.2. You can't be serious about asking for concrete examples. It forgets details after three messages. When I give it code to review it makes up mistakes that aren't there. I paid for a year under Grok 4.2, I got ripped off.

@thesoragirls You need to get money for tokens somewhere

@stepango If anything it's inverted 🤔

What does it mean for AI doomers?🤔

As soon as your team really starts to adopt AI to write code you hit the wall of “more code than eye ball time” - engineers are overwhelmed with generated code. Code reviews started to be unfair, cause it’s generated. IMO code is “disposable” asset now.