deceptiveness

The problem is that internal communications inside these companies raised concerns about the manipulativeness, and even deceptiveness of the algorithms and tactics they were using.

It is exactly because the common response to the evidence is this kind of deceptiveness that I have to point it out.

Certainly not to an Alex Jones level of deceptiveness, and to a much higher degree of public service.

Trying to claim that it was somehow the AI’s fault is sheer incompetence and/or self-serving deceptiveness.

The system may generate a body of text that a human reader may attribute to deceptiveness but there is no intent.

It's deceptiveness that makes influence or persuasion manipulative, not the tools and techniques.

I don't think it's deliberate deceptiveness, but I do think it's the rhetorical equivalent of negligent malpractice.

Those are both fair distinctions (Silent vs Notifying and the deceptiveness vs 'bad').

Right, so latent deceptiveness has to be favored in pretraining / RL.

We should expect latent deceptiveness in the same way we see reasoning laziness pop up quickly.

I assert the article would have been better focusing on the deceptiveness than bloviating about...

I'm coming around to agreeing with you about the deceptiveness of Tesla's marketing of Autopilot.

The deceptiveness of training a NN on a game whose rules are conditional probability structures and then claiming the very-good-quality conditional probability structures it finds are "World Models" is...

I'm not going to pretend it's a slam dunk "here's a paper from the esteemed Harvard Professor of Lying-is-Awesome, presenting the grand unified theory of 'deceptiveness and social skills are overlapping traits'," but it's pretty much the null hypothesis.

I think it's more of a attack on the two-faced deceptiveness that politicians usually exhibit, and the demand for "unpopular choices" is more of a demand for consistency in the face of opposition looking to score political points on a hot-button issue.

To that end, it needs to be: a) useful to be deceptive to achieve CoT reasoning progress as benchmarked in training b) obvious deceptiveness should be "selected against" (in a gradient descent / RL sense) c) the model needs to be able to encode latent deception.