Basavasagar Patil (@Neurips25)

CS PhD Student at @UCIrvine

[Neil Zeghidour @neilzegh]

Me defending my O(n^3) solution to the coding interviewer.

[Lizhang Chen @lzchen_ut]

modded-nanogpt WRs 🤝 Cautious Weight Decay

Pretty much all the @speedrun WRs since November have used CWD in some form (e.g., @varunneal's "CWD w/ schedule").

Huge thanks to everyone experimenting and sharing results — shoutout to @varunneal, @classiclarryd, @ChrisJMcCormick, @roeeshenberg, @YouJiacheng, and the other contributors.

Paper 👇 https://t.co/0qWhQnUuLv

[Basavasagar Patil (@Neurips25) @basavasagar18]

There aren't much things that break my intuitions in deep learning, but when a training run breaks it the feeling is awesome!!

[Basavasagar Patil (@Neurips25) @basavasagar18]

wandb down??

[Basavasagar Patil (@Neurips25) @basavasagar18]

@unixpickle Lol😅

[tobi lutke @tobi]

My annual MRI scan gives me a USB stick with the data, but you need this commercial windows software to open it.

Ran Claude on the stick and asked it to make me a html based viewer tool. This looks... way better.

[Basavasagar Patil (@Neurips25) @basavasagar18]

@cloneofsimo they have such different characters, but love using both of them!!

[Basavasagar Patil (@Neurips25) @basavasagar18]

So deep in ML trenches that searching for Loss Angeles instead of Los Angeles!

[Sergey Karayev @sergeykarayev]

Claude Code with Opus 4.5 is a watershed moment, moving software creation from an artisanal, craftsman activity to a true industrial process.

It’s the Gutenberg press. The sewing machine. The photo camera.

[Basavasagar Patil (@Neurips25) @basavasagar18]

coding agents are a great showcase of how unlocking small skills like using tools with high reliability can change how we interact with the system!

[flavio @flaviocopes]

--dangerously-skip-permissions

[Basavasagar Patil (@Neurips25) @basavasagar18]

"The enormous shortage of ability to compute is distorting our work, creating problems where there are none, making others impossibly difficult, and generally causing effort to be misdirected. Shouldn’t this view be more widespread, if it is as obvious as I claim?"
-Moravec, 1976

[Zhengdong @zhengdongwang]

In my annual letter this year, I really try to get at what it means to "feel the AGI."

Featuring compute, inevitability, second-order effects, travel tips, Andor, and Isaiah Berlin.

https://t.co/tnamR3EWL0

[Basavasagar Patil (@Neurips25) @basavasagar18]

The problem is what is the pre-training recipe for understanding all of these different inputs? Also can some of these "tools" can be finetuned? I fell like a universal tokenizer just a like alphabets of language is the first problem to solve

[Basavasagar Patil (@Neurips25) @basavasagar18]

a claude-code version of robotics would be a policy that has access to tools like SAM3, specialized grippers, action vision, video-models... when needed, not by default!

[Basavasagar Patil (@Neurips25) @basavasagar18]

@NielsRogge @cursor_ai they lost access to free claude-code i guess! that would sure be the reason to come back, lol

[Basavasagar Patil (@Neurips25) @basavasagar18]

@infoxiao @_arohan_ Lol, the first thing that came to my mind after reading his tweet was also that, dude thinks we are millionaires hanging out here!

[Basavasagar Patil (@Neurips25) @basavasagar18]

@AlexGDimakis which workshop was this??

[Basavasagar Patil (@Neurips25) @basavasagar18]

@HessianFree https://t.co/1zsXRAWrld I think this paper is very similar @Jaeyeon_Kim_0

[Basavasagar Patil (@Neurips25) @basavasagar18]

@HessianFree what was the previous approach??

[Basavasagar Patil (@Neurips25) @basavasagar18]

@eric_alcaide what is hard to get right?