rajan agarwal (@_rajanagarwal) / X

rajan agarwal

2,113 posts

rajan agarwal

@_rajanagarwal

coding models @openai

Joined March 2018

Pinned
rajan agarwal
@_rajanagarwal
Nov 25, 2025
in order to have research agents that can run for days, we need context compaction i used RL to have LLMs naturally learn their own 10x compression! Qwen learned to pack more info per token (ie use Mandarin tokens, prune text) read the technical blog:
rajan.sh
LLMs can invent their own compression - Rajan Agarwal
As a constrained optimization problem, LLMs can use RL to invent their own compression schemes to increase its context window.
173K
rajan agarwal
@_rajanagarwal
Apr 15, 2025
thanks claude
239K
rajan agarwal
@_rajanagarwal
Feb 11, 2025
only in waterloo
361K
rajan agarwal
@_rajanagarwal
Jun 8, 2024
the entire bee movie on github
00:00
209K
rajan agarwal
@_rajanagarwal
May 13, 2024
startups dead: sentiment analysis live meeting assistant translation apps language learning apps music/singing generation tutoring apps interview prep apps coding assistants agent interaction mental health data science live analysis wearables customer service storytelling & more
779K
rajan agarwal
@_rajanagarwal
Jun 23, 2024
gradient ascent has been achieved internally
115K
rajan agarwal
@_rajanagarwal
Jun 24, 2024
had dinner with @karpathy :) robotics, llms, travel & Fortnite??
207K
rajan agarwal
@_rajanagarwal
Sep 26, 2024
We built a startup from scratch with AI Agents! By giving agents access to Slack and custom tools using @replicate, @cohere and @v0, our simulation did market research, created branding, posted a job on LinkedIn and built a full-stack app! Read below to learn how it works.
138K
rajan agarwal
@_rajanagarwal
Oct 15, 2025
make nanochat multimodal for < $10! this evening, i trained nanochatVL: via a projection model (llava-style) between SigLIP ViT and @karpathy nanochat to extend its understanding to images it's a huge wip rn, but have a few promising results! now i can finally sleep
00:00
Andrej Karpathy
@karpathy
Oct 13, 2025
Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,
105K
rajan agarwal
@_rajanagarwal
Jun 24, 2024
MY LIFE IS COMPLETE
82K
rajan agarwal
@_rajanagarwal
Sep 11, 2025
cursor can now read ur terminal in real time! we built a subagent to watch logs, training runs, security alerts, and step in to patch code with the cursor-cli
00:00
eric zakariasson
@ericzakariasson
Sep 11, 2025
really impressed by what the people we invited built at cursor hq tonight
120K
rajan agarwal
@_rajanagarwal
Feb 20, 2024
At Stanford this weekend, we built Nexus: a graph visualization software that helps you discover meaningful relationships at events We scraped Slack data to extract interests and backgrounds, then using NLP to find similarities!
00:00
81K
rajan agarwal
@_rajanagarwal
Dec 19, 2024
we're turning @uwaterloo into a supercomputer! arceus is a cross-device distributed compute network for training large models, using model/tensor/pipeline parallelism. you can train anything, from deep neural networks to language models on the network. oss & deployment soon!
00:00
36K
rajan agarwal
@_rajanagarwal
Jun 9, 2025
a few life updates! - joining @AmazonScience agi lab this fall to work on RL postraining for web agents! - turned 20 today! will be back in sf this weekend for ai yc startup school! who’s around?
44K