4 ms·

Yes. That is my experience as well. Some people think LLM-powered models will be able to solve this issue. I think that is being overly optimistic. But maybe th

by dvrp 4y ago

Yes. That is my experience as well. Some people think LLM-powered models will be able to solve this issue. I think that is being overly optimistic. But maybe that's what we technologists have to be—especially the ones building the apps of the future. To be fair, there *are* golden nuggets of information out there. But you have to know where to look for; be an information connoisseur. Anyways, how do you cope with the current shallowness of the web?

Rastonbury 4y ago
I tend to use reddit for reviews, but need to pick through all the fake self promo, just a matter of time before it becomes as useless as Google
- ehnto 4y ago
  Especially with GPT generated reviews becoming so convincing. GPT detection is a startup idea with commercial legs I think. Reddit would want to ne able to weed them out, as not to end up with a "market of lemons" effect on the whole site's legitimacy.
joenot443 4y ago
Tragically, I’m of the opinion LLM’s are going to make this much, much worse. You can buy SEO optimized writing for a few fractions of a cent/word. When that price permanently drops to 0 (and with ChatGPT, it debatably already has), I think the resulting blogspam will start drowning out any real human signal left. Depressing stuff.
- pixl97 4y ago
  A million years from now the orginally human created Von Neumann probes are going to enter some alien solar system and ask about their cars extended warranty, then proceed to disassemble everything and move on to the next solar system to ask about theirs.
rchaud 4y ago
> Some people think LLM-powered models will be able to solve this issue. The people that have money to throw at this, tend to align their incentives with the needs of the people that pay them. So I'm not that hopeful.
ehnto 4y ago
For technology, I go straight to the docs or deep dive the codebase and just work out what's required. For information where I need a humans opinion I will often search only on reddit, and if it's a fact I'm looking up I will search only on wikipedia. There's also a somewhat hard to relay dance of keywords and specificity in language that you learn over time. "Big cat" versus "cats of atypical measured size" for a made up example. In one I expect sites with funny images of big cats, in the other I am trying to target language used in research or more analytical articles. Here is where I think language models will make search worse, because a language model will be generalizing search terms and trying to surface articles by using specific language may stop working as well.

by dvrp 4y ago

Rastonbury 4y ago

I tend to use reddit for reviews, but need to pick through all the fake self promo, just a matter of time before it becomes as useless as Google

ehnto 4y ago
Especially with GPT generated reviews becoming so convincing. GPT detection is a startup idea with commercial legs I think. Reddit would want to ne able to weed them out, as not to end up with a "market of lemons" effect on the whole site's legitimacy.

joenot443 4y ago

Tragically, I’m of the opinion LLM’s are going to make this much, much worse. You can buy SEO optimized writing for a few fractions of a cent/word. When that price permanently drops to 0 (and with ChatGPT, it debatably already has), I think the resulting blogspam will start drowning out any real human signal left. Depressing stuff.

pixl97 4y ago
A million years from now the orginally human created Von Neumann probes are going to enter some alien solar system and ask about their cars extended warranty, then proceed to disassemble everything and move on to the next solar system to ask about theirs.

rchaud 4y ago

> Some people think LLM-powered models will be able to solve this issue. The people that have money to throw at this, tend to align their incentives with the needs of the people that pay them. So I'm not that hopeful.

ehnto 4y ago

For technology, I go straight to the docs or deep dive the codebase and just work out what's required. For information where I need a humans opinion I will often search only on reddit, and if it's a fact I'm looking up I will search only on wikipedia. There's also a somewhat hard to relay dance of keywords and specificity in language that you learn over time. "Big cat" versus "cats of atypical measured size" for a made up example. In one I expect sites with funny images of big cats, in the other I am trying to target language used in research or more analytical articles. Here is where I think language models will make search worse, because a language model will be generalizing search terms and trying to surface articles by using specific language may stop working as well.