4 ms·
> I debated with Claude endlessly about this selection model, and Claude made me discard a bunch of interesting but less defensible claims. But in the end, I wa
by xyzsparetimexyz 9mo ago
> I debated with Claude endlessly about this selection model, and Claude made me discard a bunch of interesting but less defensible claims. But in the end, I was able to convince Claude it’s a good model
This is not a good way to do anything. The models are sychophantic, all you need to do in order to get them to agree with you is keep prompting: https://www.sfgate.com/tech/article/calif-teen-chatgpt-drug-advice-fatal-overdose-21266718.php https://www.sfgate.com/tech/article/calif-teen-chatgpt-drug-...
- mrefish 9mo ago> Let me know what you think by complaining that AI sucks on HN! At least you complied with the next sentence :) EDIT: Whoa, I didn't check your link before I posted. That's terribly sad. While I agree that LLMs can be sycophantic, I don't think Yegge was debating with Claude about drug use in this situation. Other references might have worked better to support you claim like this first page result when I search for "papers on llm sycophancy": https://pmc.ncbi.nlm.nih.gov/articles/PMC12592531/ https://pmc.ncbi.nlm.nih.gov/articles/PMC12592531/
- xyzsparetimexyz 9mo agoI don't think that AI sucks. I just think that it is something that is too maleable to debate with
- doug_durham 9mo agoYou need to explicitly tell it to debate you. Also you cite and example of someone using ChatGPT to discuss personal issues. We are talking about technical discussions here.