Community proposes unified 'One dataset' for open LLM training

A Reddit user suggests building a single high-quality dataset instead of coordinating decentralized training runs. The proposal responds to recent scarcity of open-weight model releases and government bans on commercial frontier models.