Sriraam

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 friend test it out by trying to add a payment method 💳 to Amazon. The cursor walks through the entire process 💬 while talking and pointing 🖱️ to the right parts of the website. Powered by Gemini 2.0 Flash (Experimental)⚡ from @Google and their live multimodal API. Shoutout to @alexanderchen for sharing the starter code that powers most of this app 🙌🔥

@kalomaze Wow

lol

Today, at the @DARPA expMath kickoff, we launched 𝗢𝗽𝗲𝗻𝗚𝗮𝘂𝘀𝘀, an open source and state of the art autoformalization agent harness for developers and practitioners to accelerate progress at the frontier. It is stronger, faster, and more cost-efficient than off-the-shelf alternatives. On FormalQualBench, running with a 4-hour timeout, it beats @HarmonicMath's Aristotle agent with no time limit. Users of OpenGauss can interact with it as much or as little as they want, can easily manage many subagents working in parallel, and can extend / modify / introspect OpenGauss because it is permissively open-source. OpenGauss was developed in close collaboration with maintainers of leading open-source AI tooling for Lean. Read the report and try it out:

Cool work like this makes RL even more exciting 🔥

Jensen name-dropped me in the keynote and posed with our belt. He has a physical belt too but they just showed the pic Intially I made fun of the 35X perf improvement being bogus, I thought it was an exaggeration of performance Turns out he was sandbagging, and perf is 50x

(1/n) Introduce Online Experiential Learning toward the era of experience. Beyond offline pre-constructed training data, models can learn online from their own deployment experience across infinite, unsimulable real-world environments. Accumulate, consolidate, self-improve 🔄

Come to Sundai and hack on RL Environments with @PrimeIntellect 🔥 hmu if you’re gonna be there! Link 👇

@maxbittker @PrimeIntellect Ofc!

@27upon2 @PrimeIntellect Let’s compare notes some time! I’m just getting started on this

This is so interesting, funny how you can just try things with the @PrimeIntellect Lab

@tm23twt @PrimeIntellect Extracting data directly from chart images but honestly I think I’ll have to move to a system where the model just gives coordinates and the harness handles pre and post processing.

@27upon2 @PrimeIntellect whachh u cookin broo

prime eval tui

Introducing FlashCompact - the first specialized model for context compaction 33k tokens/sec 200k → 50k in ~1.5s Fast, high quality compaction

Introducing Unsloth Studio ✨ A new open-source web UI to train and run LLMs. • Run models locally on Mac, Windows, Linux • Train 500+ models 2x faster with 70% less VRAM • Supports GGUF, vision, audio, embedding models • Auto-create datasets from PDF, CSV, DOCX • Self-healing tool calling and code execution • Compare models side by side + export to GGUF GitHub: github.com/unslothai/unsl… Blog and Guide: unsloth.ai/docs/new/studio Available now on Hugging Face, NVIDIA, Docker and Colab.

@tensorqt It’s a pretty leaf

let's see if anybody guesses what this is

ok guys you can draw with claude :-) sound on!! lelezhang.design/draw

@ChinmayKak We don’t get that logged in the Prime Lab, atleast now

@27upon2 How does the entropy look like

I don't like it when the model gives me hope and then crashes

Finally finished! If you're interested in an overview of recent methods in reinforcement learning for reasoning LLMs, check out this blog post: aweers.de/blog/2026/rl-f… It summarizes ten methods, tries to highlight differences and trends, and has a collection of open problems

@ElaraJordan yea I thought it can’t be cold coming from Boston but it was kinda cold

be me, move to SF > learn that startups throw millions around like it’s nothing > get over the fact that it’s actually kinda cold > be a girl > master vibe coding > meet founders > help them out on projects Do this, never have to work for anyone or think about money again.

pro tip for AI cos the main AI guy with the most power and resources is hellbent on ensuring that open models thrive and that secrets don’t stay secret and that lifelong post-training is a key ingredient of the future of AI products there is a lot of upside in leaning in here