4 ms·
In any retail webstore, or even on Spotify, YouTube, etc. all the searches are infuriating today. If I search for specific terms then I get a menu of slop to c
by ButlerianJihad 2mo ago
In any retail webstore, or even on Spotify, YouTube, etc. all the searches are infuriating today.
If I search for specific terms then I get a menu of slop to choose from. Search isn't built to serve the searcher at all. Search is meant to surface things that are promoted, hyped, advertised, or paid for in some way.
So you search Amazon for stuff and a lot of crap added on to it. Any retail site's filters are useless for filtering out what you don't want to see or casting a net for desirable features. I swear--search for "bananas" at a grocery store and you'll get results for ice cream and baby food and Minions merchandise and organic pears?
This behavior has already crept into LLM responses. I often request that they compile a list of the top-5 or top-10 according to some criteria. And it never fails that they cannot resist giving me outliers and things that don't fit by any means. Now I don't know what influences an LLM to give stupid non-matching results, but I know they do. Is RLHF already baking-in advertising and sponsorship? I wouldn't rule it out.
Case in point: the other day I was thinking of Amy Schumer but couldn't remember her name. So I struck up a convo with Copilot, and described Schumer by profession, personality, and physical attributes. Copilot first suggested a slender Black woman. Then Diablo Cody. And eventually compiled me a list of 5-7 people who were nothing like Amy Schumer. Never once even thought to mention her. Once I surfaced her name by other means, I went back and scolded Copilot, just so I could feel smug and better about myself. But it's genuinely surreal.
- prepend 2mo agoWhat physical attributes did you use. When it comes to funny or successful female comedians, I don’t think Amy Shumer would be in the top 7. Also, maybe you’re hitting an equity filter that purposely doesn’t compare race, sex, weight, etc. Recently chatgpt said it couldn’t estimate the race and age of a person in a photo because it would violate ethical norms. The photo was of me and then I tried again with a friend of mine.
- initatus 2mo agoMe: Give me a list of 30 minute podcasts ... Gpt3something: <List> The first one has 90 minute episodes! 30 minutes only! <You can listen for 30 minutes across three sessions.>