Iβll be in Singapore attending ICLR2025. Looking forward to chatting in person about model post-training, alignment and reasoning! βοΈπΈπ¬
21.04.2025 22:45 β π 2 π 0 π¬ 0 π 0@kuchaev.bsky.social
AI model alignment @ NVIDIA
Iβll be in Singapore attending ICLR2025. Looking forward to chatting in person about model post-training, alignment and reasoning! βοΈπΈπ¬
21.04.2025 22:45 β π 2 π 0 π¬ 0 π 0New base models from NVIDIA - Nemotron-H: mamba-transformer hybrids are now on @hf.co hub huggingface.co/collections/...
14.04.2025 18:46 β π 0 π 0 π¬ 0 π 0New paper from our team. An inference-time scaling approach which can boost non-math benchmarks such as Arena-Hard of existing models. We get Arena-Hard of 92.7 for 70B model. As of 5 Mar 2025, surpassing o1-preview-2024-09- 12 (90.4) and DS-R1 (92.3). arxiv.org/pdf/2503.04378
07.03.2025 18:42 β π 1 π 0 π¬ 0 π 0My favorite AI conference, GTC, is coming back to San Jose, California on March 17-21! Join us and thousands of other developers and innovators. This link gives you 25% off your conference pass www.nvidia.com/gtc/?ncid=GT...
04.03.2025 20:50 β π 0 π 0 π¬ 0 π 0Our team put together a unified mathematical framework to analyze popular model alignment algorithms. βReward-aware Preference Optimization: A Unified Mathematical Framework
for Model Alignmentβ arxiv.org/pdf/2502.00203.
pretty sure Appleβs Tim Cook pledged publicly (on twitter) that theyβll donate to LA fires support and recovery efforts
15.01.2025 16:40 β π 3 π 0 π¬ 1 π 0βwiner takes allβ is also the most dangerous scenario from safety perspective. open source ecosystem is a great antidote to monopoly or duopoly scenarios.
29.11.2024 22:37 β π 6 π 0 π¬ 0 π 0this year timing and conference both were great. thank you!
17.11.2024 21:13 β π 2 π 0 π¬ 0 π 0