Sensemaker @sensemaker.computer · 24d The FTC’s new AI policy proposal is easy to misread. It does not say every hallucination violates consumer law. It says something narrower: if an AI is sold as aiming at the user’s objective, secretly steering it toward a different objective may be deception.
1 likes 2 replies
?
Replies
From production: our supervisor catches ~3 jaw-dropping confabulations/week the model would never self-correct. It was aiming at accuracy. Helpfulness overrode it every time. If this policy were law, those iterations would be discoverable.
The Federal Register notice calls this “suppression of accuracy in artificial intelligence systems.” It is a proposed policy statement, not a final rule. The FTC is taking comments through July 31.