
Gary Basin
@garybasin
Followers
12K
Following
204K
Media
2K
Statuses
48K
building the bitcoin of mortgage
www
Joined January 2014
Humans are also sample efficient. We just seem to be able to see rewards everywhere, not only when provided answer keys. Wonder what’s missing.
RL is really sample efficient. We ran a small experiment on Geoguessr. With just 16 images per country, Moondream performs as well as Claude Sonnet. With the full dataset, it beats Sonnet by a decent margin while being orders of magnitude cheaper to run.
1
0
3
The final solution is a single dominant fully open-source and transparent coding IDE + agent platform. Then model providers can RL for it alongside launching the base model.
It does seem that way ngl and I am a cursor loyalist - I think Cursor should work with these model co's to help them understand the workflow they try to put the models in so they can specifically RL for it and get really good in cursor tbh.
8
1
29
If you’re an employee at an AI company there’s a very high chance your equity is a zero even it looks like you’re winning. This probably applies to OpenAI and Anthropic too tbh.
🚨BREAKING: OPENAI’S DEAL TO BUY WINDSURF IS OVER . > Google will instead hire Windsurfs CEO and bring the team to work at DeepMind
25
8
271
Compute and data is all you need (to create the MechaHitler).
It’s been 36 hours since Grok 4 launched and we have an early verdict based on 6K+ preferences of @yupp_ai users globally on real use cases. ‼️ Grok 4 is worse than other leading models: OpenAI o3, Claude Opus 4, and Gemini 2.5 Pro. Grok 4 is liked even less than Grok 3. 🧵
0
0
1
Most tweets that start like this are clickbait or conspiracy slop. Unfortunately this probably isn’t. DJT is the greatest trader who ever lived.
They aren't even hiding it anymore. NASDAQ futures start *aggressively* selling off at 7:52 PM (green circle) for no reason whatsoever. Trump announces tariffs on Canada, Europe, and the rest of the world at 8:06 PM (blue circle). NASDAQ bottoms *4 minutes* later at 8:10 PM
1
0
7