Mastodon Feed
jonny@neuromatch.social ("jonny (nonvenomous)") wrote:
I genuinely try these tools, I use them with all the recommended usage patterns, I try to do all the magic context management tricks, they sometimes hit the slot machine jackpot and get something right, but the rest of the time they just make doing what I know should be done 10x as expensive and confusing.
There is a new brand of post in the anthropic subreddit, people being amazed at how good the advice fable gave them. The reason they cite is that fable pushed back on them heavily and arrived at some Socratic solution. But if you ask them "OK so was the advice actually good" (e.g. did investing advice make you money, did relationship advice save your relationship, etc.) The answer is crickets. The new pattern for convincing people of model efficacy seems to just be negging them, and it also seems to be an effective enough product change to sell through the IPO