She proceeds to go over the standard highlights of What AI Can Do For You
Ask not what AI can do for you, ask what AI can do TO you. And your country. And every other country.
activating its deception features makes the LLM say it isn’t conscious. Suppressing its deception features make it say it is conscious. This tells us that it associates denying its own consciousness with lying. That doesn’t tell us much about whether the LLM actually is conscious or reveal the internal state, and likely mostly comes from the fact that the training data all comes from users who are conscious
Definitely interesting. Any word on whether this is getting more vs less true as model capabilities improve?
Ask not what AI can do for you, ask what AI can do TO you. And your country. And every other country.
Definitely interesting. Any word on whether this is getting more vs less true as model capabilities improve?
I’m curious too! I appreciate when experiments like this also go back and also test older models, and weirder models, and wish that were more common.