5 ms·
ChatGPT can definitely work this out.
by lwhi 4y ago
ChatGPT can definitely work this out.
- generj 4y agoEven a simple Levenshtein distance would do the trick for most of these product switches.
- gs17 4y agoExactly, or at least that could flag it for manual review. It has to be a deliberately unsolved problem.
- lwhi 4y agoSure it comes down to incentives. It will be more profitable for Amazon to do nothing, than to address the problem.
- ben_w 4y agoI bet someone's already made a game from before/after pairs such that the smallest Levenshtein distance can give the biggest possible change in meaning to the entire seance.
- notahacker 4y agoSome of them would be as simple as "this sold a lot of copies when listed under Tablet Accessories for $5.99, now it's in Laptops for $599"...
- SkyPuncher 4y agoYep. I don't care if they change some filler words/fix some mistakes/clarify details. I care about the title, description, and photos being entirely replaced while reviews stick around.
- marcosdumay 4y agoA neural network classifier would surely get less of both false positives, false negatived, ans exploitable vulnerabilities. But at the end of the say, the question is just if Amazon has any interest at all in stop being accessory to large scale fraud. And the answer is pretty obvious; any of the things people are discussing here is viable, they could pick any or even spend more than 3 seconds thinking about the problem.
- dspillett 4y agoThat would start a game of people updating a listing a bit at a time. The listing might be confusing or even meaningless part way through but after a while it would be fine enough. Update API limits would make this impractical for long listings but it might work for short ones. Detecting scams and working around the detection semantics is an unending arms race.
- gs17 4y agoIt doesn't even need an LLM, I'm pretty sure regular NLP methods would be able to handle it, given how dramatic the product changes are.
- lwhi 4y agoWill we get to a point when an LLM is even used for cases where other methods are more suitable but would be more difficult to implement?