Suraj Srinivas

@Suuraj

ml researcher @ bosch research / trying to understand deep learning

Sunnyvale, CA

suraj-srinivas.github.io

Joined June 2009

1,193 Following

1,168 Followers

Replies Reposts Media

Pinned
Suraj Srinivas
@Suuraj
Jun 22
📢New ICML 2026 position paper: "Explainability research should prioritize foundations over ad-hoc methods." Despite countless explainability techniques (incl. feature attribution, concept-based methods, SAEs), explanations rarely influence real workflows. Why? We believe this
3
Suraj Srinivas
@Suuraj
Nov 4
True in both science and business
Bojan Tunguz
@tunguz
Nov 3
The gap between perception and reality is the ultimate arbitrage opportunity.
Suraj Srinivas
@Suuraj
Oct 28
"In-context tokenization" feels like magic! Behaviors reminiscent of a truly intelligent machine. Great work @cartesia!
Albert Gu
@_albertgu
Oct 28
I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data. Language
2
Suraj Srinivas
@Suuraj
Oct 23
Hypothesis: the allure of "open science" may be at odds with startup incentives. Many academic-minded startups genuinely want to do open science and build in public, which is great. But often, small early wins get widely shared while later failures stay private. Without meaning
Lucas Beyer (bl16)
@giffmana
Oct 23
On the zeroth order lm training stuff: It's cool that people try stuff instead of only reading and dooming. But i feel the need to curb the enthusiasm on this one a little. Two things: (1) methods in the genetic-algo and fwd-diff categories, which imo these fall in, is
Quote
spicylemonade
@spicey_lemonade
Oct 22
WE DID IT! We can now pretrain transformers without backprop and even beat backprop+Adam on validation loss with insane speeds. Now the big question is scale. It’s Ironic, we have reached a breakthrough In AI capability due to inspiration from a safety blog
3
Suraj Srinivas
@Suuraj
Oct 21
This looks like an incredible result! Together with @Recursive_SI's announcement, we seem to be heading into the era of completely automating model training This is a great direction IMO and has the potential to greatly democratize AI development
Behnam Neyshabur
@bneyshabur
Oct 21
We formally announced Mirendil less than 4 months ago. Today, I'm excited to share an important milestone for: Pretraining at scale without a pretraining team. Mirendil's mission is to solve Recursive Self-Improvement (RSI) and use it to democratize frontier AI development,

Suraj Srinivas

Suraj Srinivas’s posts