Rushil Chugh

[levent @__alpoge__]

Please welcome to the world a beautiful new geometric object, to do with a problem i’ve always loved. claude really contains multitudes:D Does S^6 admit a complex structure?

Yup

[Rushil Chugh @chugh_rushil]

@iiiichigo_chan Slop

[Rushil Chugh @chugh_rushil]

@alokbishoyi97 Hey, worked on post training at Amazon, would love to be a part

[Rushil Chugh @chugh_rushil]

@alokbishoyi97 Hey, love to be a part, could you dm please

[Rushil Chugh @chugh_rushil]

@alokbishoyi97 Hey, dm'ing about the group

[clem 🤗 @ClementDelangue]

We just crossed $100M annual run-rate. I know many AI companies are capturing much more $$$ these days, but still proud of the milestone!

Maximizing short-term revenue has never been our priority. In fact, we're proud to manage to store and serve hundreds of petabytes of models and datasets while keeping HF free and open-source for 97% of our users. As a platform, we’re happy to hopefully create orders of magnitude more value for the community than what we capture. To me, that’s the very definition of a platform.

And it has helped us build one of the most loved platform in tech, with network effects, a defensible position and a sustainable business which is quite unique in AI.

Many many thanks to all the community members for building with us, we wouldn't be anywhere without you! Can’t wait for what’s next, especially as more companies start to see the value of open and local AI! Next milestone $1B?

[Rushil Chugh @chugh_rushil]

@loganthorneloe Perhaps a section about rl and control theory and it's implications on robotics ?? It's often notoriously hard to get the reward functions correct right

[Rushil Chugh @chugh_rushil]

@DimitrisPapail new eval just dropped, explain backdprop without going dumber

[Nathan Lambert @natolambert]

I don't really want to have to go to bat against Anthropic, but they've just been unnecessarily antagonistic to all of China, then not so subtly to open weight models, and now more broadly open AI research. What's next on the list?

[Rushil Chugh @chugh_rushil]

@yoheinakajima @ActiveGraphAI @harvey yep, the nice thing was that during synthesis the outputs felt so much more easier to trust cus of the trace. Not a universal memory win, but if work is around controls, obligations, gaps etc. activegraph fits great !!

[Rushil Chugh @chugh_rushil]

@MichelIvan92347 @harvey Thankyouu !!

[Rushil Chugh @chugh_rushil]

1/ I ran a small ablation on agent memory.

Question: if you preprocess a large document corpus into a searchable memory layer, does it help an agent do downstream work?

Used @harvey's Legal Agent Benchmark because legal is a brutal proxy for document-heavy knowledge work.

[Rushil Chugh @chugh_rushil]

16/ Broader lesson:

Before adding memory, run the ablation first.

Keep model + judge fixed.
Compare against no-memory and raw search.
Inspect task shape.

Post: https://t.co/hKGw8wUPlM

[Rushil Chugh @chugh_rushil]

15/ Big caveats:

single runs
judge variance not measured
criterion pass rate, not all-pass success
best-observed table is diagnostic, not deployment performance
framework + preprocessing are entangled

Still, the pattern was useful.