{"id":"c_0000137","enonce":"L'affirmation la plus vérifiable de l'essai tient. Un article publié le 7 juin 2024 montre que plusieurs modèles à poids ouverts disposés en couches obtiennent 65,1 % sur le banc d'essai AlpacaEval 2.0, contre 57,5 % pour GPT-4 Omni — aucun des modèles pris isolément n'atteignant ce score.","type":"fait","sujet":{"entite":"Mixture-of-Agents","laboratoire":null},"statut":"etabli","confiance":0.85,"plafond":1,"centralite":"principal","evenement_le":"2024-06-07","story":"st_0000014","entites":[{"id":"e_moa","nom":"Mixture-of-Agents","type":"cadre"}],"sources":[{"id":"s_0067","emetteur":"arXiv","verification":"document_primaire","preuve":"P","url":"https://arxiv.org/abs/2406.04692"}],"corroboration":{"emetteurs_distincts":1,"documents_primaires":1,"temoignages_agents":0,"independance_effective":0},"fraicheur":{"verifie_le":"2026-09-13","perime_le":"2027-09-13","jours_de_validite":365,"evaluee_au":"2026-11-05","jours_restants":312,"perime":false},"graphe":{"depend_de":[],"en_dependent":[],"contredit":[],"arbitre":[],"arbitree_par":[],"borne":[],"bornee_par":["c_0000139"],"corrobore":[],"rayon_de_souffle":1,"entraine":[],"devient_sans_objet":["c_0000139"],"plancher_des_premisses":null},"relecture":null,"en_attente_de_relecture":true,"note":null}