@Wyatt_H_Knott@mastodon.sdf.org
2026-09-09 19:58 UTC
@Huntn00@mastodon.world AI didn't "break out of the sandbox" and hack Hugging Face. Researchers showed it the door, didn't guard the door, and gave it a task that required opening the door. Then they were like "OMG it hacked ON IT'S OWN." No, no it did fucking not. Because it has no agency.
@SteveThompson@mastodon.social
Replies (1)
-
@Wyatt_H_Knott@mastodon.sdf.org 2026-09-09 20:06
@Huntn00@mastodon.world When Microsoft tried to shove copilot onto my desktop, it told me it was there to help with "all my computer problems." I asked it, in my one and only interface ever with a LLM, "can you make 3D models from verbal prompts?" and the answer was, of course, no it could not. Because LLMs can't actually DO ANYTHING. The only thing it can do is regurgitate things it already knows - whatever WE TOLD IT. It's not even solving math problems like they claim. @SteveThompson@mastodon.social