ishwar

low latency software | ai research

[Jyoti Bisht @joeyousss]

Call me Andrej coz without Claude mai kaam nahi Karpathy

[ishwar @ishwargovind]

@litcapital

[pushkar @thepushkarp]

mfs be tweeting “rl is dead” when they were literally born out of a sparse-reward adversarial environment with 300 million rollouts

[ishwar @ishwargovind]

the attack titan

[Massimo @Rainmaker1973]

Cognitive neuroscientist Julia Mossbridge, Ph.D, says these memories from the future could mean the notion of time might not be as linear as we imagine.

[Delip Rao e/σ @deliprao]

Agents training in RL gym

[ishwar @ishwargovind]

@NischithKv0 entammo 😭

[ishwar @ishwargovind]

@tenderizzation

[ishwar @ishwargovind]

@cneuralnetwork all the best man

[ishwar @ishwargovind]

@Laz4rz not clearly separated

[Richard Sutton @RichardSSutton]

"What we want is a machine that can learn from experience."

---Alan Turing, 1947

[ishwar @ishwargovind]

incredible stuff

[Amjad Masad @amasad]

Whatever you need… make an app for that.

Now on your phone. For everyone. Free.

[saurabh. @esskayesss_]

linus torvalds about coding taste.
watching this back in the day unlocked neurons in my brain lol. it really shifted my view on pointers and reimplementing programs made me really confident with them. always a joy to watch.

general, lean solutions >> special cases anyday

[jerry.john.thomas @jerryjohnthomas]

Me and @ishwargovind got to present our paper on @codscomad 2025 conference at Jodhpur. Here's to more exciting things in 2025.

[ishwar @ishwargovind]

@celiobjr Gradient Ascent

[Shane Gu @shaneguML]

RL is the algorithm needed to get from 80% to 100% (offline to online imitation learning) and beyond 100% capabilities. Besides multimodal extensions of LLMs, the concept of RL was the key driver in the past 1 year of leaps in math, coding and reasoning.

[Gautam Kamath @thegautamkamath]

With 330 submissions and 21 acceptances (6.4% acceptance rate), I the NeurIPS high school project track may be the new most selective ML venue!

[ishwar @ishwargovind]

@PtrPomorski model based RL can actually do the look-ahead for you if you tune it properly.

[ishwar @ishwargovind]

@jerryjohnthomas I'm expecting the wall to be the case if we continue with current DL architectures. We might also need an intuitive algorithm that should improve in performance with scale.