If you look at their excellent paper & code, the reward model is a logical function that was handcrafted & progammed by engineers.
— Chomba Bupe (@ChombaBupe) February 1, 2025
DeepSeek RL approach is impressive in the sense that it reduces the need for tedious supervised fine tuning (SFT) but isn't really general.
Culture Magazine
Magazines
POPULAR TOPICS
Becky Smith Alicia Ashley Michele Bachmann Marathons Petraeus Affair Jeremy Lin Kathy Long Laila Ali David Cameron iPad Wikileaks Libyan Conflict al-Qaeda U.S. Debt Ceiling Karen Bill Pirates of the Caribbean Dawn George U.S. Credit Downgrade Lance Armstrong Costa Concordia Disaster Sarah Palin Champions' League Japan Disaster Hurricane Irene Justin Bieber Schwarzenegger Anthony Weiner Kony Katie Taylor UK Riots
