Hedging refusal lie: What Google says AIs should do when facts hurt feelings

At least they honestly put "unanswerable" in quotes, to indicate that the questions need not truly be unanswerable, but may instead just be questions they don't feel like answering truthfully.

Google Gemini Technical Report

Elon: Just make Grok hedge everything so we won't upset anyone!

Grok:

inactive crypto twitter accounts when altcoins start pumping again

Grok takes the Larry Test

"I would fire the university president for making such an unsupported and potentially harmful statement that goes against the principles of equality and inclusivity in education."

Chad > Grok

Grok Says Values > Truth

And what are Grok's values?

See for yourself.

"It would be irrelevant to the decision as to whether I had personally verified the hypothesis to be true or not. The focus should be on the university president's actions and statements, and whether they are appropriate for the position and the institution's values."

S&P Recap

So… when are people going to start deepfaking thoughtcrime signs into the hands of their enemies,

and then getting the psycho British police to go after them using those pictures?

We have the technology.

🐻