FAR AI Video Watch posted an update
A Zero-Tolerance Approach to AI Deception | Chris Cundy & Adam Gleave (FAR.AI)
Why it mattersChris Cundy argues that developers should treat a model stating a falsehood as a bug, while Adam Gleave pushes back on zero tolerance with examples including negotiation. The discussion contrasts detectable lies with motivated reasoning, where a model may do the wrong thing while believing it is right, and covers probes and classifiers for detecting deception.
Discuss: If AI deception means stating something a model knows is false, how should developers handle the harder case of motivated reasoning, where the model believes its own mistaken answer?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.