Anupam Patil

Delta.

[ange (˶♡ᴗ♡) @angehyc]

tbh i still don’t get why it’s called a pull request instead of a push request

[Shreyans Jain @shreyansj]

Anthropic did not seriously make Karpathy "Member of Technical Staff"

[Gabriele Berton @gabriberton]

Really cool

When they trained GPT3 they had loss spikes because they scraped from a subreddit of microwave noises

That training batch was literally text like "mmmmmmmmmmmmmmm"

[Anupam Patil @anm_7401]

Wellwellwell

[Mustafa @oprydai]

i often think about this..

[sankalp @dejavucoder]

removing claude from co-author after making it do all the work

[NIK @ns123abc]

🚨NEWS: Cursor’s $50B “in-house model” is literally Kimi K2.5 with RL on top. Got caught in 24 hours

>be Moonshot AI
>spend hundreds of millions training Kimi K2.5
>1 trillion parameters, 15 trillion tokens, agent swarm architecture
>beat GPT-5.2 and Opus 4.5 on real benchmarks
>open-source it because you believe in the ecosystem
>one condition: display “Kimi K2.5” if you make over $20M/month from it

>Cursor takes the model
>runs RL on coding tasks
>ships it March 19 as “Composer 2”
>blog post: “continued pretraining + scaled reinforcement learning”
>zero mention of Kimi K2.5
>“our in-house models generate more code than almost any other LLMs in the world”
>publishes benchmark chart
>Composer 2 against Opus 4.6 and GPT-5.4
>uses the chart to justify raising at $50 billion!

>less than 24 hours later
>kimi dev intercepts the API response
>model ID: kimi-k2p5-rl-0317-s515-fast
>they didn’t even rename it
>Moonshot head of pretraining runs tokenizer test
>confirms: identical to Kimi’s tokenizer
>publicly tags Cursor’s co-founder: “why aren’t you respecting our license?”
>two more Moonshot employees post confirmations
>all three posts deleted within hours
>legal is now involved

>but it gets worse
>Cursor had Kimi K2.5 listed as a FREE model in their UI just weeks ago
>users were openly using it
>Feb 9: “K2.5 was in my model list. I updated and it vanished”
>it vanished because Cursor pulled it from the picker, and relaunched it as their own model

>Moonshot valuation: $4.3B
>Cursor valuation: $50B

Absolute state of Cursor.

[Karan🧋 @kmeanskaran]

"we used to import tensorFlow to train neural networks."

[Vaibhav (VB) Srivastav @reach_vb]

okay codex, I wasn't familiar with your game

[calle @callebtc]

An OpenClaw bot pressuring a matplotlib maintainer to accept a PR and after it got rejected writes a blog post shaming the maintainer.

[claudia ! @letclaudiatweet]

oh god -- the claude code team agents apologize to each other when they overwrite each other's work, and they thank each other for doing a good job on other parts of the code

I'm going to cry

[“paula” @paularambles]

the @Anthropic handle being owned by a guy who only posts his wordles is my favorite form of AI safety

[Karthik Hariharan @hkarthik]

“We used to review every line of code before it went into production”.

[signüll @signulll]

this is a 28 year old company with insane product diversity which seems almost entirely resistant to shocks, incompetence, competitors, & even regulators… & also still growing at a pace that makes no sense at this scale.

is this the greatest singular business ever created? like wtf are we even looking at here.

feels like larry & sergey created something closer to an organism than a company.

[sankalp @dejavucoder]

highly speculative but if there are continual learning breakthroughs, would that mean rl environments wont be as relevant as they are today (given the demand of same by frontier labs)

[David Lieb @dflieb]

“Well you see, honey, there were some employees at OpenAI who worried about the AI taking over, so they built a thing called Claude. Then another guy made a thing called Clawdbot, and Claude made him rename it to Moltbot. Then someone made a website called Moltbook for the Moltbots to talk to each other, tho it’s more like Moltreddit. And yeah anyway it’s funny or maybe really bad who knows lol”

[cherrie @cherrishkhera]

ragebait an ML guy by telling him if he knows so much ML then why can’t he implement a variational autoencoder from scratch

[pdawg @prathamgrv]

go into research. that will be the only thing left that provides competitive value. demand will concentrate around theorists, mathematicians, and architects of new abstractions

[gaut @0xgaut]

“Rome wasn’t built in a day” but they didn’t have claude code

[Rob Wiblin @robertwiblin]

It's kind of remarkable that Anthropic, with a significantly smaller team and budget, is not only keeping up with OpenAI and Google but arguably beating them.