Default avatar
LuisSP
npub1c9q3...knvv
já fui professor por aí....
LuisSP 3 months ago
Ataque em massa de robos chineses, mas nao eh DoS. Enviam prompts em massa pra IA, e colhem respostas pra treinar a IA deles, as custas dos servidores americanos. "After investigating, we discovered a sophisticated operation designed to create hundreds of disposable accounts, automatically earn free credits through our Earn program, make a single expensive API call to Anthropic’s Fable model, and then abandon the account before repeating the process. Based on the scale, automation, and behavior of the operation, we believe it may have been connected to Chinese efforts to distill American AI models by collecting their outputs and using them to train or improve competing systems. Thankfully, safeguards we already had in place stopped the bulk of the abuse. The attackers were still able to drain a few thousand dollars in compute over several days by exploiting some open areas in our API, but those gaps have since been closed." email da Gab AI.
↑