Requests for Research 2.0

OpenAI published Requests for Research 2.0, a list of seven unsolved ML and RL problems with starter tasks.

The prompts span parity training for LSTMs, Snake and multiplayer Snake RL environments, and self-play setups.

Other challenges include parameter averaging in distributed RL, transfer learning across Atari with generative models, and linear-attention transformers.

The release is meant to accelerate progress in reinforcement learning, sequence modeling, and more practical long-context model design.