EleutherAI
- Type
- Civil society organisation
- Lieu
- États-Unis — United States (non-profit research institute, distributed/online-first)
- Dernière vérification
- 2026-09-22
- Prochaine vérification
- 2027-03-21
Pas encore traduit — affiché en anglais.
Comment les joindre
- General enquiries
- Discord — the actual organisation; research happens in public channels
Ce canal est conservé uniquement comme référence. Sa vérification est absente, date de plus de 180 jours, ou la période de contact n'est pas ouverte actuellement ; ne vous y fiez pas avant une nouvelle vérification.
- GitHub
- Homepage
Ce qu'il fait
Grassroots-origin open research lab. Trains and releases open-source LLMs (GPT-NeoX, Pythia), maintains the LM Evaluation Harness used across the field, and researches interpretability (interpreting across training time), eliciting latent knowledge, alignment, and non-text modalities. Runs almost entirely in public on Discord.
Évaluation franche
The most genuinely open organisation in this entire list. An unaffiliated person with technical skill can join the Discord today and be contributing to alignment-relevant research within weeks. Caveat: it is a research community, not a complaints channel, and its open-release stance means it is not a natural home for a restrict-frontier-models concern.
Préoccupations couvertes
Mixed and openness-centric: the core thesis is that open models and public interpretability tooling are necessary for anyone outside the large labs to audit AI. Touches alignment (ELK, interpretability) but is not an x-risk organisation; it is frequently in tension with the pause/restriction wing.
Canal gouvernemental
Modest — EleutherAI has filed comments in US policy processes (notably on open models) and its researchers are cited on open-weight policy, but it has no advisory seat.