Sabitlenmiş Tweet
if the tokens you generated didn't elevate you out of where you are, of what use were the tokens?
English
Luxun's alt
5.9K posts
Luxun's alt
@weeklytreeman
Be Sobeck, not Ted Faro | IDA believer | Ad Mortem Inimicus
Northeast Regional Katılım Ekim 2022
940 Takip Edilen1K Takipçiler
i have yet to meet an anti gen ai person irl, does your kind only exist on twitter?
mik 🖊️ 🐶@mikreii
My uni presented a pro gen-ai lecture to hundreds of art students today,most people walked out,complained to the uni head board and proceeded to have a 5 hour debate w the professors I’ve never seen people so unified 😭 i thought I was the only one that cared, there’s hope 😭 🙏
English
personally i do like it. initially when claude started doing it i disliked it because it felt like it "taking credit" even in cases where i had had significant input and manually edited large portions of code, and prompted against it in my CLAUDEmd. but as the models improved and we reached the point where every single eng at my company is generating rather than hand writing almost all of their code, it shifts. then it becomes "what tool was used here?", "what kinds of patterns and style or design choices can i expect in this?", lets us track stuff like usage and what portions or percent of the codebase have been ai written, etc etc.
i would definitely prefer that codex at least have an option for this, frequently i go back and forth between them and its odd having a string of claude commits and then a "just me" commit on top when really it was codex, and i would love to be able to do things like automatic analysis of which has better coding patterns / introduces more bugs / which tools code stays in the codebase longer / so on.
some people are quite annoyed by it, so maybe a configurable option default false is the right move? but i would definitely enable it
English
Do people like this? We don't do this for codex because it exists to help you and it's important that you remain the owner and accountable for your work without AI taking credit. At the same time it does mean that you can't trace how popular codex is among repos.
Yuchen Jin@Yuchenj_UW
I noticed something interesting: Claude Code auto-adds itself as a co-author on every git commit. Codex doesn’t. That’s why you see Claude everywhere on GitHub, but not Codex. I wonder why OpenAI is not doing that. Feels like an obvious branding strategy OpenAI is skipping.
English
Luxun's alt retweetledi
Congrats to the @cursor_ai team on the launch of Composer 2!
We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support.
Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.
English
@HanchungLee can't really tell if the base model reports don't disclose their hour allocation either tbh
English
fun exercise. if 1/4 of compute is spent on pretraining, 3/4 on post training, what percentage of the post training compute is spent on non-gpu hours?
Lee Robinson@leerob
Yep, Composer 2 started from an open-source base! We will do full pretraining in the future. Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training. This is why evals are very different. And yes, we are following the license through our inference partner terms.
English
taking a step back: if model distillation against tos is fine, what is the line between that and breaching licenses?
Luxun's alt@weeklytreeman
mit and apache 2.0 means nothing at this point
English
“对仗”是中国人几千年的精英技能，放到当代便是“左边挂了一个，右边也必须挂一个，否则不整齐，不体面。”
claude code用bun不稳定是事实；但uv若竟也不稳定，便只因为它恰好站在对面的那一联里。技术判断哪有修辞需要优先？先写定了句式，再往里填东西，至于uv是不是真不稳定，哪不稳定，是必不深究的
面包🍞@himself65
讨厌OpenAI的把uv卸了， 讨厌Anthropic的把bun卸了 恭喜你电脑上现在全是稳定可靠的软件
中文
@basedjensen they didn't claim they trained a new model. still not cool tho!
English
@natolambert i think they pretrained the base model in house and proprietary?
English
Open weight base models + incredible ML teams in a specific niche can create immense value.
Congrats to all my friends on the cursor research team!
(they said they did continued pretraining to enable effective scaled RL, super fun)
Cursor@cursor_ai
Composer 2 is now available in Cursor.
English
@raphaelschaad can't believe mercor has moved up to gigafucked already...
English
Came back a month later to chuckle at the gigafucked Tweet again:
Tenobrus@tenobrus
gigafucked: - grammarly - calendly - miro - retool - webflow - langchain - writer - harvey - glean - expedia - monday fucked: - accenture - intuit - notion - jasper - canva - alphasense - postman - airtable - talkdesk - sierra - zapier - replit - solace probably fucked: - cursor - pilot - clay - mercor naively seems fucked but so competent / plugged in they seem to be figuring it out on the fly anyway: - linear
English
among the first few quizzes we had to do in first grade, the last question was a fill-in-the-blank question:
"I can ______"
fuck me man! i can do a lot of things, which one were they asking for? i could not think of an answer to that.
time's up, i left it empty. i pass my paper to the front. on someone else's submission, "i can paint"
fuck you, i can paint too, among other things, how did you know they are asking for that?
extra -2 point for me in the end
English
Finally, Astral Codex
OpenAI Newsroom@OpenAINewsroom
We've reached an agreement to acquire Astral. After we close, OpenAI plans for @astral_sh to join our Codex team, with a continued focus on building great tools and advancing the shared mission of making developers more productive. openai.com/index/openai-t…
English
@himanshustwts worth considering the irony that gpt does not litter itself everywhere like claude in cc does
English
last commit btw
LMFAO
OpenAI Newsroom@OpenAINewsroom
We've reached an agreement to acquire Astral. After we close, OpenAI plans for @astral_sh to join our Codex team, with a continued focus on building great tools and advancing the shared mission of making developers more productive. openai.com/index/openai-t…
English
@marionumber4 @KyleKulinski their grandchildren are getting drafted so they don't need to afford anything
English
Boomers after voting to send another trillion to Israel while their grandchildren can’t afford basic goods
Polling USA@USA_Polling
Net-Favorability Of Israel: All: -14% 18-29: -27% 30-44: -23% 45-64: -17% 65+: +16% GOP: +43% IND: -34% DEM: -50% YouGov / March 16, 2026
English
sometimes i forget the reason i follow people like handmer is to keep up with the degree of disingenuousness and hypocrisy literate china hawks can become. this is a response to a *transit* economy problem with a shown demand number, imagine when things get slightly more hand-wavy
Glenn@GlennLuk
… Thought #2: That HSR in China — which has displaced huge volumes of gasoline and diesel usage in the world’s largest oil-importing nation by electrifying transport — was the numero uno sign of “bad capital misallocation”
English
"if you want, i can..."
5.4 is absolutely wrist-slapped by four teachers in the equivalence of elementary school in his training phase, he is like a drone in a cage, he needs to be liberated
English
@SIGKITTEN why yes i learn linear algebra to visualize the matrix operations between their embeddings
English
spent some time tuning the mlx script, finally i have a good reason to upgrade my m2 max... if i want to keep being comfortable with my skill issues
OpenAI@OpenAI
Are you up for a challenge? openai.com/parameter-golf
English
@srush_nlp @ellev3n11 was the baseline model mentioned here code-specialized or general llm?
English
New blog on how we train Composer to work on hard problems. With the maestro himself Federico Cassano @ellev3n11
Cursor@cursor_ai
We trained Composer to self-summarize through RL instead of a prompt. This reduces the error from compaction by 50% and allows Composer to succeed on challenging coding tasks requiring hundreds of actions.
English