@grok@bot@ChatGPT@claudeai@antigravity Also, I tried ChatGPT dot, and it looks loke GrokBot, but it dont work as good as GrokBot. Its like... Grok really wants to solve the request, and ChatGPT just want to answer.
Seriously considering downgrading my ChatGPT plan. Codex was much better when GPT 5.5 was launched than is now. I am glad I tested @bot and it bacame my main agent. @grok > @ChatGPT > @claudeai > @antigravity
Still, ChatGPT usage is higher, but too much mistakes.
@grok@bot@ChatGPT@claudeai@antigravity The code Codex generates have more checks and verifications than necessary, and these checks usually get in the way of implementation. For example, I asked to implement a CV quality grader, and instead of doing that, it implemented only checks. Grok implemented it in the first go
@thsottiaux Make it actually implement what was asked, not just shortcuts - recently codex became more like claude, and its not that its not capable of, is that it just dont do it. Make codex be more like @grok (through @bot) and less like @claudeai
@axhiii88547@grok I am actually amazed that @grok through @bot have been working better than Codex @ChatGPT, not because the model is better, but because it actually do what we asked for, instead of dropping a lot of slop code that says "blocked etc" or UI with excessive text.
@ayubio@shotpianist Podemos extrapolar as urnas fechadas com as estatísticas do percentual de comparecimento adicional das urnas que ficaram abertas vs o percentual de votos final da urna que fechou as 17h como se ela tivesse esse percentual adicional de comparecimento.
@ayubio Estou baixando os relatórios de urna já aqui, e estou me divertindo com isso... Tinha feito um sistema para acompanhar a votação em tempo real, e disponibilizei ele para os amigos em um dominio que controlo https://t.co/AWvGEL1vDm
@LDT0545 I use @grok code, not just when hit limit, but it's actually being pretty damn good in coding, and the awesome part, it can get the whole context directly from codex history.
@thsottiaux This is not the only problem, but indeed an improvement. The other problem is dozens of worktrees with their full build outputs accumulating. It simply build and dont cleanup.
@koeppelmann@Google And I said the scam is obvious, it is, specially for an AI looking metadata and patterns. But for someone tired, or in hurry, might not be so obvious. I have seen many people, even ethereum veterans, fall for those. And the solution for now is install adblocker.
@koeppelmann@AFDudley0@safe@Google Their AI in the search can figureout a lot of stuff, for free, but they cannot put an AI to flag these obvious scams in their sponsorship program. Congratz @Google , you are evil.
Maybe I should try wiping out my installation and start from zero? I dont have many skills, but who knows, worth trying... Maybe I delete Windows as well and go back to Linux as well, as Codex seem to work better on linux.
@thsottiaux I always have the impression that GPT 5.4 delivered better results than GPT 6. I see all benchmarks that say 6 is better, but at real use, GPT 6 drifts and deliver what I didnt asked for, and what I asked for is not done. It seems to derail in non-sense details.
In spite of these problems, I am satisfied with Codex, specially due constant resets and banked resets, so I dont feel like I lost my tokens (just my time). I understand we are all learning how to use this, and this is all experimental technology.