soul_grafitti
Joined 13 years ago
Joined 13 years ago
Disappointed in the VC move but I don't have any complaints at the moment. I have had good experiences with Cloudflare for other services so feel predisposed to them.
> Anthropic is also quick to link this new research to its previous work on sycophancy, noting that “sycophantic validation” is “the most common mechanism for reality distortion potential.”
I commented on Ars that I see the same sort of mechanism at work in politics. So is the really an AI problem or a case of AI surfacing inherently human characteristics? It's a sincere question.
I don't think I'm alone here when I say I frequently gain understanding of things by writing code to model them. I know I understand something when I can write it in code. So after ingesting the lesson about a topic I'll see if I can write code to do what I just learned. I have spent a lot of hours stuck at a juncture of not being entirely sure my code, which works on the simple examples, is also working in the complex ones. The difference in results may not be evident. Am I troubleshooting my understanding or my code? I need a known result to actually know.
This is one place being able to copy something from your textbook, for example, paste it into ChatGPT and ask questions is a huge benefit. ChatGippity doesn't always get it right but usually the interchange is much more useful than being stuck.