And just now I had to interrupt Codex amidst a 48 minute turn, where I asked for a rather simple / well-defined implementation, not some outlandish "build me a 1M b2b saas" type of task.
It's keep getting "stuck" during exploration, without progressing any further.
/fast is unusable as websocket fails to connect.
I suppose the issue might be related with the session length; I'm getting responses at normal pace again once I started up a new session and gave Codex the previous session id and asked it to continue from where we left off. I've been working in previous session for several days now. So the problem might be solved, but it was unpleasant experience nonetheless.
@thsottiaux Hey Tibo, sure the tokens must flow but codex connectivity has gotten worse for sure.
I'm keep getting reconnecting... n/5 and codex is taking over 30 min for a sufficiently simple task which would've taken at most 12 minutes few days ago on 5.5 xhigh.
Wow, this tweet went very viral!
I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs.
So here's the idea in a gist format: https://t.co/NlAfEJjtJV
You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.
이런 것도 오래가진 않겠지만 말이다.
Harness는 말 그대로 말을 다루기 위한 마구인데, 이 말은 신비롭게도 시간이 지날수록 더 빠르고 강력하고 똑똑해지는 성질을 가지고 있고, 심지어 기수의 성격에 맞춰 자신을 바꾸기도 하기 때문에 곧 기수가 마구를 필요로 하지 않는 때가 올 수밖에 없다.
어제 이 내용관련 녹화를 싸이오닉 대표님과 찍긴했는데, 새로운 세상 새로운 세대의 가치관을 보여주는 시작점이라고 본다. 인류의 모든 지식과 표현을 능가하는 super intelligence 의 존재자체가 모든 저작권의 super set 인데 그들의 모든 산출물이 저작권 위반일텐데 어쩌란 말인가. 과��의 관점으로 해석이 불가능하다.
Leaked code를 가져다 올리고 AI로 언어를 바꿔 작성하게 하곤 “clean room”이란 그럴듯한 단어를 가져다가 force push로 git history를 갈아치우며 아무 문제가 없다고 하는 것에 대한 짜증은, 그 행동의 적법성은 둘째치고
이런 후안무치함이 보상받는 현 시대에 대한 염증일 것이다.
천재적이긴 하죠, 유출 코드 그대로 올려서 초기 스타 붐업 실컷 한 다음에 코드 파이썬으로 갈아엎고 클린룸이라 주장하는 어뷰징 솜씨가요
이딴 짓거리를 지금 천재 개발자니 뭐니 잘한다 잘한다 띄워주면서, 기사 첫머리에는
> “유출된 코드 활용, 윤리의 문제” 비판도
https://t.co/TnbDaXrutL
단순하게는 중앙은행이 관리하는 CBDC를 줄기로 DeFi 등 Private 섹터의 작은 가지들이 뻗어나가는 형태가 예상된다.
https://t.co/j8mskawKdD
다만 지금껏 암호화폐의 특징이었던 수많은 알트코인을 위시한 heterogeneous한 생태계는 등장하지 않을 확률이 높아 보인다.
People make decisions with emotions, not logic, but that's actually rational if "rationality" means a predictable relationship between an input and a corresponding output.
@LLMJunky@thsottiaux thanks, maybe xhigh is being too greedy now that the model has gotten smarter and possibly that is why it's going off track? i'll def try high for the time being
@thsottiaux gpt 5.4 (xhigh, without /fast) is keep straying off the reasoning path, and ends up producing response not to the last user message but to the one before that.
@thsottiaux gpt 5.4 (xhigh, without /fast) is keep straying off the reasoning path, and ends up producing response not to the last user message but to the one before that.
it feels awesome; overall a much better "writer", understands code and docs holistically, has the both of claude-like familiarity and gpt-5.x-codex like precision.
except that there's weird bug where it goes off reasoning trajectory and responds to the second last user message instead of the last message.
not sure if i'm the only one experiencing this (using Codex App btw)
@thsottiaux so i'm resending the message until it gets right (which definitely doesn't help my usage limits) didn't have this happen yesterday and earlier today, only started seeing this since this afternoon :/
@thsottiaux but in the second screenshot, i asked codex to give me a summary of our conversation in the current session so far, and it started to reason towards that, only to stray off the path and generate response about the parity between the eng and kor docs