Sriraam
@27upon2
Followers
1K
Following
9K
Media
228
Statuses
2K
building @decodetool. playing with RL envs. prev @Harvard
Boston, MA
Joined July 2016
Introducing Gemini Cursor ✨ – a second multimodal AI cursor for your desktop that's open-source and free! Link below 👇 This experiment 🧪 reimagines how we interact with our computers because visual cues 👀 help us make sense of what we see on a screen. In this demo, I had my
🔥 @Google Gemini 2.0 Flash is crazy good at pointing. I was over engineering before but now I'm just gonna bet on model capabilities. This is a demo of an AI cursor explaining a diagram on @tldraw with just a prompt and an image. Streaming is also simple with @vercel AI SDK.
32
109
1K
I’m Vibe RLing with TOML files
Hosted Training Create your environment, configure your training run, and we handle the rest. No worrying about managing infrastructure, GPUs, or low-level algorithms. We’re launching with agentic RL, and adding support for SFT and other algorithms in the near future.
1
4
30
Excited to launch Lab — the full stack for training your own models Unifying RL environments, hosted training, and evals into one platform Going from research to optimized model without infra headaches
Introducing Lab: A full-stack platform for training your own agentic models Build, evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab.
12
15
198
If we want models to behave well in the real world, we need to train and evaluate them in something closer to the reality. Releasing carla-env, an open source embodied environment that gives models access to engineering scale physics simulation. All details are in blogpost:
22
34
315
If you have a harness, irrespective of programming language and tech stack you can now train and eval models without worrying about GPU infra. There is no excuse not to RL anymore
Introducing Lab: A full-stack platform for training your own agentic models Build, evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab.
2
3
40
Over the past few weeks in private beta, more than 3,000 RL runs were completed by individuals and companies from around the world. Starting today, we’re opening it up to everyone.
3
4
96
Introducing Lab: A full-stack platform for training your own agentic models Build, evaluate and train on your own environments at scale without managing the underlying infrastructure. Giving everyone their own frontier AI lab.
111
251
2K
> Spent the last few weeks exploring active context management with @PrimeIntellect hosted RL beta. > By enforcing strict memory wipes to test reasoning, INTELLECT-3 demonstrates clear grokking phase transitions. (1/2)
1
3
51
reward went up but after looking at rollouts found out that i didn't account for some edge cases so need to stop and do thorough testing locally
1
0
9
we just released prime-rl v0.4.0 highlights: * Bring your own algorithms, advantages and loss can be extended via plugin without touching prime-rl code. Useful for researcher that want to plug their own recipe on top of a powerful async rl engine for large scale moe * Multi
5
8
154
The LLM Engineering Roadmap. If you want to start today, here's the roadmap👇 1️⃣ LLM Foundations Start by understanding Python and LLM APIs and how they work. Learn prompt engineering, structured outputs, and tool use. ↳ Python/Typescript Basics ↳ LLM APIs ↳ Prompt
21
239
1K
2025 RL: get dataset, build env, find gpu, ssh and setup os deps, fix issues, OOM while u sleep, pissed off, get fat gpu, run works, upload rollouts to HF or download, write scripts to analyze, update rewards, repeat it all, hope it works, tweet 2026 RL: get dataset, build env,
3
4
108
Announcing slowmo: Slow down, pause, or speed up time of ANY web content. Try the demo and download the extension here: https://t.co/pqjYt8vsqb Debug animations, learn from cool demos, and make games easier or harder. Available as an extension: https://t.co/dtKoiLpsVx
11
14
164
got a dummy RL run with the prime hosted lab teaching a model to draw diagrams on a whiteboard i was worried there would be issues cuz the env is spinning up a react app for the harness but it just worked. awesome work @manveerxyz @willccbb and the team 🔥 next up is synthetic
2
3
32
and you can use your product as the infra for RL and SFT. here I'm using @tldraw's react app to do evals and RL on open source models
the infra that enables you to A/B test models or prompts is basically the same infra that lets you do reinforcement learning
0
0
6
Gonna make Intellect-3 better at drawing with prime-rl. It seems better than Qwen3-30B based on my vibe eval. Opus 4.5 is obv the best. Making a curriculum and rubric for this is gonna be so fun lol. Finally found a fun use of the Lab @willccbb Images: @PrimeIntellect
0
0
10
