GPT-Commander

Author of HEC-Commander Tools and RAS-Commander library. Engineer. Modeler. AI enthusiast. See my open source repos for HEC-RAS modeling and automation work

[GPT-Commander @GPT_Commander]

@tysonhutchins_ @rohanvarma Remote control has been broken for me since 4.7 dropped.

[GPT-Commander @GPT_Commander]

@bcherny @theo @benhylak Bro y’all have to do better, remote control has been broken for nearly 2 weeks.

I don’t give a f about new features, fix the ones you’ve already added

[GPT-Commander @GPT_Commander]

@SubhamHasNoLife @samhogan Grep killed RAG. Last year. Handily.

[GPT-Commander @GPT_Commander]

@turbo_xo_ @MatthewBerman Codex is slow and a bit regarded if you aren’t specific enough, but it does pick up on stickiest problems and figures them out. Anthropic’s models are much better at task management and actually doing things.

[GPT-Commander @GPT_Commander]

@paulhshort @thsottiaux I use CLI’s every day for real work. I hold no equity in either competitor, this is just my consumer opinion.

[GPT-Commander @GPT_Commander]

@thsottiaux @daniel_mac8 Awesome, I’ve been waiting for nearly 6 months for most of the recent features. Recently reactivated the more expensive subscription but still lots of work to do….

[GPT-Commander @GPT_Commander]

@johnnydamacha @runaway_vol @chadeyecom I don’t know why people glaze 5.4, it’s terribly slow and just as brittle as previous iterations. I want to like it, but I wonder how people make these statements as fact. It’s like they don’t actually use the model for anything important.

[GPT-Commander @GPT_Commander]

@headinthebox LLM expert, not AI expert. It’s funny that in the age of language models we still can’t be specific!

[GPT-Commander @GPT_Commander]

@steipete @Voxyz_ai My first instruction was to never download skills from the internet. I think people rightly assume it’s a huge malware vector, and don’t report. Honestly you should just move to a codex review gate to improve trust.

[GPT-Commander @GPT_Commander]

@unclebobmartin It sounds like you really need to look at Hooks! They help immensely with exactly the type of behavior you are describing. You can force the model to loop/self-verify until complete. It’s not perfect, but it helps a lot with context drift.

[GPT-Commander @GPT_Commander]

@milan_milanovic It’s hilarious that they thought otherwise. They got greedy and overindexed on technology they don’t understand….

[GPT-Commander @GPT_Commander]

@dkundel @gajesh @thsottiaux @0thernet https://t.co/G62riAURKA

[GPT-Commander @GPT_Commander]

@AaronHenray @thsottiaux @0thernet IME Claude still reliably completes tasks with minimal steering. Especially long running tasks that rely on features within the CC harness. Codex is more powerful but requires more steering. Codes is a more powerful model with a more chaotic jagged edge and a weaker harness.

[GPT-Commander @GPT_Commander]

@AaronHenray @thsottiaux @0thernet IME Claude still reliably completes tasks with minimal steering. Especially long running tasks that rely on features within the CC harness. Codex is more powerful but requires more steering. Codes is a more powerful model with a more chaotic jagged edge and a weaker harness.

[GPT-Commander @GPT_Commander]

@thsottiaux @markchen90 @0thernet I don’t know what you guys are working on, but I use both Codex and CC regularly and I have never had this experience. Just feels like Coke vs Pepsi debates. Everyone convinced they are superior because of their subjective preference.

[GPT-Commander @GPT_Commander]

@JonahBlake Not even close LOL

[GPT-Commander @GPT_Commander]

@unclebobmartin It’s not your imagination! I only use Codex when I need a second opinion.

[GPT-Commander @GPT_Commander]

@proctorzt @joelgrus @Austen Likely already did for this to work…..

[GPT-Commander @GPT_Commander]

@Austen @joelgrus You’ve got other settings you’ve already changed, like whitelisted tool calls. Your tweet doesn’t make sense, try again on a fresh machine.

[GPT-Commander @GPT_Commander]

@Austen None of this works without the yolo flag though? You can’t override the approval with a prompt. Not sure if alpha or clickbait