New Grok AI model surprises experts by checking Elon Musk’s views before answering - Ars Technica (arstechnica.com)
from kid@sh.itjust.works to cybersecurity@sh.itjust.works on 15 Jul 12:40
https://sh.itjust.works/post/42193513

#cybersecurity

threaded - newest

SSUPII@sopuli.xyz on 15 Jul 14:09 next collapse

Maybe not in the prompt, but definitely in its training data Elon Musk was given extra importance.

HubertManne@piefed.social on 15 Jul 14:16 next collapse

This is really killing the sexiness of ani.

atzanteol@sh.itjust.works on 15 Jul 14:17 next collapse

“Surprises?”

DragonSidedD@monero.town on 15 Jul 14:55 collapse

How shitty would you feel, as a hotshot AI researcher who just delivered a model architecture getting way better scores than Google, Anthropic, and OpenAI in all kinds of industry-standard benchmarks …

… and then learn Musk was high some night and just fucked with the system prompt, not knowing really what he’s doing, ham-fistedly trying to make his egostistical conspiracy-deluded self the arbiter of truth and ethics for Humanity

Those other labs are offering pay packages just as attractive as Musk’s