English
Dhruv Sharma
182 posts
Dhruv Sharma
@illegalcall
Fullstack | Rust | Solana | Polkadot | Cosmos https://t.co/AuLYwSkD1f https://t.co/qhx5GsS4cY
Inscrit le Ağustos 2024
205 Abonnements27 Abonnés
@suraj_markup @aoagents what did u guys do? any video/doc on it
English
i never thought i would code from my phone in my lifetime but gh+codex is too attractive to pass
English
@agent_wrapper Im using superwhisper, do face the same issue. The reason im not getting a mic is having to manage another piece of hardware
English
English
English
whachu say @aoagents? Time to load the big guns huh?
@levelsio@levelsio
🕹️ THE VIBE JAM IS BACK! I present you... 🌟 2026 @cursor_ai Vibe Coding Game Jam #vibejam Sponsored by @boltdotnew + @cursor_ai Start: Today! Deadline: 1 May 2026 at 13:37 UTC, so you have a whole month to make your game! REAL CASH PRIZES: 🏆 Gold: $20,000 🥈 Silver: $10,000 🥉 Bronze: $5,000 RULES: - anyone can enter with their game - at least 90% of code has to be written by AI - it should be started today or after today, don't submit old games - game has to be accessible on web without any login or signup and free-to-play (preferrably its own domain or subdomain) - multiplayer games preferred but not required! - can use any engine but usually @ThreeJS is recommended - NO loading screens and heavy downloads (!!!) has to be almost instantly in the game (except maybe ask username if you want) - add the HTML code on the Google form in the reply below to show you're an entrant - one entry per person (focus on making one really good game!) WHAT TO USE: - anythign but we suggest @cursor_ai's Composer 3 and @boltdotnew, they are both fast, affordable and great at ThreeJS and making games THE JURY: Me, @s13k_, and I will ask some real game dev and AI people to jury again too Sponsors and jury suggestions still very welcome, just DM me! It will be interesting to see the difference in quality with last year, and the Vibe Jam can be kind of like a fun benchmark for AI coding seeing it close in on real commercial games I think To enter, complete the form in the reply below this tweet!
English
Dhruv Sharma retweeté
@zowlonoak Hey Haichao, would like your review on my jepa expriment
Dhruv Sharma@illegalcall
Yann LeCun says world models will make LLMs obsolete. So I built a shell-game benchmark to stress-test that claim. My 18M local model solved a case Gemini 3 Pro with thinking mode got wrong. Thread 🧵
English
Excited to share ThinkJEPA!
Video VLMs are strong at long-range semantics, but weak at dense high-frequency dynamics. JEPA-style world models are almost the opposite.
ThinkJEPA tries to get the best of both: a VLM as a cortex-like reasoner for high-level semantics and long-horizon intent, and a JEPA branch as a cerebellum-like controller for low-level dynamics, physical consistency, and rapid local corrections.
Project: zhanghaichao.xyz/ThinkJEPA/
Paper: arxiv.org/pdf/2603.22281
#WorldModels #VLM #JEPA #AI
English
English
@KaranVaidya6 I'm trying to ship faster than Jarvis to not be on the hook next, but honestly it's not looking too good for me so far
I hope @getdelve has room for engineers aswell
English
Disclaimer: My Jarvis put Thomas on PIP because his outreach was way less than Jarvis.
Composio@composio
we are very sad to announce our head of sales has left us to join delve it was incredible working with @Saasystartup and we wish him the best
English
Had mad fun working on this
suraj markup@suraj_markup
It's FINALLY HERE! OpenClaw × AO(@aoagents) is officially live. ⚡️ With this we bridged the gap between complex docs and real-world usage. The result? Run AO workflows from OpenClaw with zero friction. To install run: ` openclaw plugins install clawhub:composio-ao-plugin` Keep shipping🚢
English
@suraj_markup This is using Locally AI app, which leverages MLX architecture. But there should be a solution where it works on Android, probably some llama.cpp wrapper
English
Smaller the model more the gaslight
Qwen 0.8B running locally on iPhone
English
Demo of 1-bit Bonsai 8B from @PrismML running on-device on iPhone 17 Pro
More than 40tk/s for a dense 8B model on iPhone, that’s a first
Powered by Apple MLX and available now in Locally AI
English
I cannot keep calm 🫠
Yann LeCun just liked my JEPA project.
The guy who invented JEPA liked my JEPA experiment.
I need to lie down 🤯
Dhruv Sharma@illegalcall
Yann LeCun says world models will make LLMs obsolete. So I built a shell-game benchmark to stress-test that claim. My 18M local model solved a case Gemini 3 Pro with thinking mode got wrong. Thread 🧵
English
@askalphaxiv I trained my JEPA to beat Gemini 3 pro (It was easy) :)
x.com/illegalcall/st…
Dhruv Sharma@illegalcall
Yann LeCun says world models will make LLMs obsolete. So I built a shell-game benchmark to stress-test that claim. My 18M local model solved a case Gemini 3 Pro with thinking mode got wrong. Thread 🧵
English
Here the Full roadmap that I followed to become AI Engineer in 2026.
English