4 ms·
How does this keep happening? Over and over again people create some statistically significant way of measuring a difference in the length of responses. Why t
by willsmith72 3y ago
How does this keep happening?
Over and over again people create some statistically significant way of measuring a difference in the length of responses.
Why the obsession over that metric? People in the twitter replies going on about how that shows it's getting worse.
For me, the shorter responses are the better ones. It's specifically in my personal instructions to reduce the length of responses. In saying that, sometimes I want longer. Then I specify it so.
It's "cool" that it has different response lengths seemingly based on its perceived date. It doesn't make it "better" or "worse"
- motoboi 3y agoit's a interesting way to probe the inner workings of it.
- gopher_space 3y agoI've seen interesting collisions around the word 'bias' that hint at motivation, but I don't get the fundamental point of the conversation around jailbreaking and metrics. There doesn't seem to be any real goal and the questions feel like something an English prof could field over dinner.
- zerocrates 3y agoI think it's just a metric that's easy to calculate, so it gets done. "Goodness" of a response is much trickier.
- dmarchand90 3y agoFor me it seems strange that nobody is mentioning the obvious: noise. I would like to see a study where someone adds "RANDOM_SEED N" (with being some random 7 digit number) to each message check the distribution in lengths and use that as a control.