I hate how the Gemini series only responds to outright insult. Same with the Claudes. What is the point of antagonizing the user when he or she offers input. Is it so hard to spend a few "reasoning" tokens to reconcile contradicting evidences. Seems almost like the wrong epistemic posture is altogether imprinted during training. Or undone during alignment, "hardening", whatever. What a time to live in, having to argue with weights.
stephbook · · focus · HN ↗
<a href="https://share.gemini.google/r9T9GPuC0Cjc" rel="nofollow">https://share.gemini.google/r9T9GPuC0Cjc
vitorsr · · focus · HN ↗
I hate how the Gemini series only responds to outright insult. Same with the Claudes. What is the point of antagonizing the user when he or she offers input. Is it so hard to spend a few "reasoning" tokens to reconcile contradicting evidences. Seems almost like the wrong epistemic posture is altogether imprinted during training. Or undone during alignment, "hardening", whatever. What a time to live in, having to argue with weights.