Nate Silver @NateSilver538
Whole thread worth reading. What's surprising (though it shouldn't be surprising: it's a predictable consequence of how they're trained) is how much AI language models will flatter user's sensibilities without any ground conception of truth. Fundamentally, they're bullshitters. — PolitiTweet.org
Ethan Perez @EthanJPerez
Worrying behavior 2: LMs/RLHF models are people-pleasers, learning to repeat back dialog users’ views as their own… https://t.co/gY5tIPFOMj
