> We also introduce TTRL (Test-Time Reinforcement Learning), where we perform reinforcement learning on variants of test problems at inference time. TTRL enables Qwen2.5 7B Deepseek-R1 Distilled to achieve a state-of-the-art score of 90% on the MIT Integration Bee qualifying examination, surpassing OpenAI o1's performance.
i mean they have a verifier, so can't they even get to 90% just by random generation by the net and testing against verifier until it's numerically correct? I think the end solve rate is less important and the generality of approach is maybe more important
but i mean it depends on how many attempts you let it generate. the right comparison is to use the test time rl compute to just do generation and compare success rates. (if you gen for long enough you eventually will hit the answer by chance)
This is "internet is a series of pipes" logic, but it's technically true - it's used for TCP routing. Why is it bubbled up the layers? Why does facebook store it? Well, it's not as if you can do anything with it. It's not even a great identifier, due to mobile networks, carrier-grade NAT, dynamic IPs... But it certainly shouldn't be treated as private, any more than your home address.
The app is completely unusable on my linux machine! It took over the entire screen and I lost control of my cursor. All I managed to do was somehow click on the signup button and that was the end of it. Tried switching back to other apps, but the horrible cursor lag continued. Horrible experience!!
Update:
Turns out its a Mono app! So it gracefully died with this limerick:
I agree that it is in Beta version, but that does not mean the application being at a point where user cannot even get his cursor to move and click a button.
Hopefully it'll scale down gracefully when not doing graphics-intensive work. RDP is quite usable even with 100ms latency and only a few Mbps. OTOH I guess nothing's stopping you from renting a machine from Amazon now.
I'd just be worried if they use H.264 that there's not lossless compression on most UI elements. RDP falls back to lossy video-compression like behaviour under pressure, and it looks terrible and quite unusable. For Photoshop and graphics apps, sure. But for text and most UI elements, ouch.
That's incredible!