Base Labs
An AI safety research team incubated by Baseten, focused on security research for open-source weight models. The team is dedicated to developing and publishing safety methodologies for training and monitoring open-source models.
Timeline (last 90 days)
Base Labs announced an AI safety collaboration with Hugging Face and Goodfire targeting open-source weight models
Base Labs is a research team incubated by Baseten earlier this year
Base Labs will develop and publicly release a methodology for training and monitoring open-source models
Open-source model safety should have safety properties embedded at the training stage, and relying solely on post-training RLHF alignment is insufficient.
The technical roadmap, delivery timeline, and evaluation benchmarks officially disclosed for this collaboration have not yet been published.
All Facts (4)
Base Labs is a research team incubated by Baseten earlier this year
50%UnverifiedOpen-source model safety should have safety properties embedded at the training stage, and relying solely on post-training RLHF alignment is insufficient.
50%UnverifiedBase Labs announced an AI safety collaboration with Hugging Face and Goodfire targeting open-source weight models
60%UnverifiedBase Labs will develop and publicly release a methodology for training and monitoring open-source models
50%