4 ms·
On the other hand, some AI generated content can be pretty good, informative, and useful. This is in the hands of its prompt's creator and in the quality of the
by OutOfHere 2y ago
On the other hand, some AI generated content can be pretty good, informative, and useful. This is in the hands of its prompt's creator and in the quality of the model. It has to be done with care to ensure quality, avoiding falsities and hallucinations.
- simonw 2y agoPart of my personal definition of slop is that it's unreviewed. If someone uses ChatGPT to help them write and iterate on a well researched essay about a topic and verifies that the details are correct, I don't see that as slop. People with English as a second language are getting a huge boost from these tools. That's great!
- Bluestein 2y ago> unreviewed This.- Unreviewed, or downright unsupervised. And, unsolicited. Perhaps, also, "unmarked".- PS. Perhaps also, "low quality".-
- OutOfHere 2y agoYou are presupposing that it's not possible for GPT to write a well researched essay, which shows a limited understanding of how to use it well for research. For targeted essays, the audience is a single person, and so there is no reason for someone to iterate on GPT's work; it is sufficient to review it.
- quantified 2y ago> If someone uses ChatGPT to help them write and iterate on a well researched essay about a topic and verifies that the details are correct, I don't see that as slop. Where is the presupposition? The parent comment seems to match your sentiment.
- OutOfHere 2y agoThe presupposition is that one has to write the essay or iterate upon it (using GPT). It is that GPT won't do it all for them. With a sufficiently developed workflow, I believe GPT can do it all. I guess we were thinking of different applications. Yes, a human always has to review the output, but not necessarily to micromanage or edit it further, sometimes only to approve or reject the claims.
- simonw 2y agoThat's why I emphasize "unreviewed". It is an uncontroversial fact that LLMs make mistakes. If a human being has reviewed the output of GPT and is willing to stake their personal reputation on that content being "correct", then I'm happy to read it. If nobody has done that then it's slop and a waste of my time. (Unless of course I requested that information from GPT myself, in which case it's on me to review it before sharing it with others.)
- Slyfox33 2y agoNo one has even come closer to ensuring quality and avoiding fake information given by LLMs. No one even knows if its possible to do.