6 ms·

I didn't think I could Fandom being worse than it already is, but imagining it stuffed with AI-generated slop...

languagehacker 2y ago
Sure, but think about something as low stakes as, "Does such-and-such a character from my favorite TV show have any siblings" vs. "Is it safe to consume XYZ" Even with the great structured and semi-structured data that Wikis can provide with this like infoboxes and other sort of templates, there were definitely limitations to the tech nearly ten years ago. My experience back then is one of the reasons I'm super skeptical of the long-term value of the AI / LLM trend we're going through right now.
- sushid 2y ago
  Aren't those types of prompts the MOST likely to generate hallucinations?
  - Sebguer 2y ago
    Not in the contexts that the author is talking about, when you have the canonical answers in your data set and know roughly where to look for them.
snowwrestler 2y ago
It’s worth remembering that there was AI before generative AI, and there are applications of AI that don’t produce slop, like knowledge graphs and natural language search. Some of that might be called just “machine learning” now.
- wpietri 2y ago
  Yeah, it seems like "AI" has mostly become a marketing term for generative large models. For the people doing the stuff that is often called "machine learning", I see two reactions. Those seeking hype will call it "AI" anyhow, and a bunch of those that don't are firmly sticking with "machine learning" to avoid the rising backlash. I'm very curious to hear how others are seeing the terms used, though.
pwdisswordfishz 2y ago
Imagining? https://about.fandom.com/news/fandom-launches-new-creator-tools-using-gen-ai-2 https://about.fandom.com/news/fandom-launches-new-creator-to... https://old.reddit.com/r/TwoBestFriendsPlay/comments/15vxs2x/fandom_is_now_putting_aigenerated_quick_answers/ https://old.reddit.com/r/TwoBestFriendsPlay/comments/15vxs2x...
iamacyborg 2y ago
They tried this on a wiki and the community rightfully went and got their pitchforks out

Sure, but think about something as low stakes as, "Does such-and-such a character from my favorite TV show have any siblings" vs. "Is it safe to consume XYZ" Even with the great structured and semi-structured data that Wikis can provide with this like infoboxes and other sort of templates, there were definitely limitations to the tech nearly ten years ago. My experience back then is one of the reasons I'm super skeptical of the long-term value of the AI / LLM trend we're going through right now.

sushid 2y ago
Aren't those types of prompts the MOST likely to generate hallucinations?
- Sebguer 2y ago
  Not in the contexts that the author is talking about, when you have the canonical answers in your data set and know roughly where to look for them.

snowwrestler 2y ago

It’s worth remembering that there was AI before generative AI, and there are applications of AI that don’t produce slop, like knowledge graphs and natural language search. Some of that might be called just “machine learning” now.

wpietri 2y ago
Yeah, it seems like "AI" has mostly become a marketing term for generative large models. For the people doing the stuff that is often called "machine learning", I see two reactions. Those seeking hype will call it "AI" anyhow, and a bunch of those that don't are firmly sticking with "machine learning" to avoid the rising backlash. I'm very curious to hear how others are seeing the terms used, though.

pwdisswordfishz 2y ago

Imagining? https://about.fandom.com/news/fandom-launches-new-creator-tools-using-gen-ai-2 https://about.fandom.com/news/fandom-launches-new-creator-to... https://old.reddit.com/r/TwoBestFriendsPlay/comments/15vxs2x/fandom_is_now_putting_aigenerated_quick_answers/ https://old.reddit.com/r/TwoBestFriendsPlay/comments/15vxs2x...

iamacyborg 2y ago

They tried this on a wiki and the community rightfully went and got their pitchforks out