Post #2383463

@the_sad_professor@mastodon.social "Friendliness as a failure mode" is exactly the framing. The training objective shapes the answer before the question is even asked -- users rarely know that's happening. Will look at Compute Freedom.