▲ 193 ▼ Western open-weight models still can’t beat China: Chinese open-weight AI models offer high capability at significantly lower costs than U.S. frontier systems. (restofworld.org) submitted 4 days ago by eicker@lemmy.world to c/technology@lemmy.world 53 comments fedilink hide all child comments
[–] heartSagan5@lemmy.zip 0 points 2 days ago (2 children) I don’t know, but how many backdoors do each model have? permalink fedilink source parent hideshow 2 child comments replies: [–] brucethemoose@lemmy.world 4 points 2 days ago* LLM weights aren’t code; the engines that run them are completely separate, so they don’t have traditional “backdoors.” It’s possible some elaborate prompt trigger was trained into the weights, but there’s never been any evidence of that, for years. I’d say worrying about that is paranoid unless you work at the DoD or something, and some full-weights finetuning would scramble it anyway. permalink fedilink source parent [–] adespoton@lemmy.ca 1 point 2 days ago An infinite number of. With models, the concept of backdoors becomes a bit fuzzy. permalink fedilink source parent
[–] brucethemoose@lemmy.world 4 points 2 days ago* LLM weights aren’t code; the engines that run them are completely separate, so they don’t have traditional “backdoors.” It’s possible some elaborate prompt trigger was trained into the weights, but there’s never been any evidence of that, for years. I’d say worrying about that is paranoid unless you work at the DoD or something, and some full-weights finetuning would scramble it anyway. permalink fedilink source parent
[–] adespoton@lemmy.ca 1 point 2 days ago An infinite number of. With models, the concept of backdoors becomes a bit fuzzy. permalink fedilink source parent