3 ms·

If the hardware changes significantly and those sites don't exist in the future wouldn't that mean gemeni would degrade in quality because it has nothing to pul

If the hardware changes significantly and those sites don't exist in the future wouldn't that mean gemeni would degrade in quality because it has nothing to pull from?

andy81 7mo ago
We've all tried to ask the LLM about something outside of its training data by now. In that situation, they give the (wrong) answer that sounds the most plausible.
- PacificSpecific 7mo ago
  That's definitely been my experience. I work with a lot of weird code bases that have never been public facing and AI has horrible responses for that stuff. As soon as I tried to make a todomvc it started working great but I wonder how much value that really brings to the table. It's great for me though. I can finally make a todomvc tailored to my specific needs.
  - ctoth 7mo ago
    I'm not sure what sorts of weird codebases you're working with but I recently saw Claude programming well on a Lambda MOO -- weirder than that?
    - PacificSpecific 7mo ago
      I had to Google that haha. It's in that realm but more complex. I do plan to repeatedly come back and try though. Just so far it hasn't been useful.
- NoMoreNicksLeft 7mo ago
  Once or twice, for me it's deflected rather than answer at all. On the other hand, they've also surfaced information (later independently confirmed by myself) that I had not been able to find for years. I don't know what to make of it.
- visarga 7mo ago
  > In that situation, they give the (wrong) answer that sounds the most plausible. Not if you use web search or deep report, you should not use LLMs as knowledge bases, they are language models - they learn language not information, and are just models not replicas of the training set.
hydrogen7800 7mo ago
Right, that success story is only because there was "organic" (for lack of a better term) information from an original source. What happens when all information is nth generation AI feedback with all links to the original source lost? Edit: A question from AI/LLM ignorance- Can the source database for an LLM be one-way, in that it does not contain output from itself, or other LLMs? I can imagine a quarantined database used for specific applications that remains curated, but this seems impossible on the open internet.
- black_puppydog 7mo ago
  That's exactly why text written before the first LLMs has a premium on it these days. So no, all major models suffer from slop in their training data.
- bigthymer 7mo ago
  > Can the source database for an LLM be one-way, in that it does not contain output from itself, or other LLMs? I think, for public internet data, we can only be reasonably confident for information before the big release of ChatGPT.
- nsvd2 7mo ago
  Yes, people have likened pre-LLM Internet content to low-background steel. If in the hypothetical future the continual learning problem gets solved, the AI could just learn from the real world instead of publications and retain that data.
- nprateem 7mo ago
  One reason why Google made that algorithm to watermark AI output
esperent 7mo ago
Presumably companies will still provide manuals.
- PacificSpecific 7mo ago
  If that was adequate then wouldn't there not be supplementary material? Results vary of course. I have some very wonderful synthesizer manuals.
- SiempreViernes 7mo ago
  It'll be a single sheet of paper with a QR code that redirects to a canned prompt hosted at whichever LLM server paid the most to the manufacturer for their content.
roxolotl 7mo ago
Yea so I’ve had an issue getting video output after boot on a new AMD R9700 Pro. None of the, albeit free, models from OpenAI/Google/Anthropic have really been helpful. I found the pro drivers myself. They never mentioned them. Thats not to say AI is bad. It’s great in many cases. More that I’m worried about what happens when the repositories of new knowledge get hollowed out. Also my favorite response was this gem from Sonnet: > TL;DR: Move your monitor cable from the motherboard to the graphics card.
elictronic 7mo ago
This then becomes the hardware manufacturers problem. If their new hardware fails for to many users it will no longer be purchased. If they externalize their problem solving like so many companies, they won't be able to gain market share. This creates financial incentives to pay companies running the new version of search. Your thinking of this as a problem for these companies, when in reality it is a financial incentive.
visarga 7mo ago
> because it has nothing to pull from? Chat rooms produce trillions of tokens per day now, interactive tokens, where AI can poke and prod at us, and have its ideas tested in the real world (by us).

If the hardware changes significantly and those sites don't exist in the future wouldn't that mean gemeni would degrade in quality because it has nothing to pull from?

andy81 7mo ago

We've all tried to ask the LLM about something outside of its training data by now. In that situation, they give the (wrong) answer that sounds the most plausible.

PacificSpecific 7mo ago
That's definitely been my experience. I work with a lot of weird code bases that have never been public facing and AI has horrible responses for that stuff. As soon as I tried to make a todomvc it started working great but I wonder how much value that really brings to the table. It's great for me though. I can finally make a todomvc tailored to my specific needs.
- ctoth 7mo ago
  I'm not sure what sorts of weird codebases you're working with but I recently saw Claude programming well on a Lambda MOO -- weirder than that?
  - PacificSpecific 7mo ago
    I had to Google that haha. It's in that realm but more complex. I do plan to repeatedly come back and try though. Just so far it hasn't been useful.
NoMoreNicksLeft 7mo ago
Once or twice, for me it's deflected rather than answer at all. On the other hand, they've also surfaced information (later independently confirmed by myself) that I had not been able to find for years. I don't know what to make of it.
visarga 7mo ago
> In that situation, they give the (wrong) answer that sounds the most plausible. Not if you use web search or deep report, you should not use LLMs as knowledge bases, they are language models - they learn language not information, and are just models not replicas of the training set.

hydrogen7800 7mo ago

Right, that success story is only because there was "organic" (for lack of a better term) information from an original source. What happens when all information is nth generation AI feedback with all links to the original source lost? Edit: A question from AI/LLM ignorance- Can the source database for an LLM be one-way, in that it does not contain output from itself, or other LLMs? I can imagine a quarantined database used for specific applications that remains curated, but this seems impossible on the open internet.

black_puppydog 7mo ago
That's exactly why text written before the first LLMs has a premium on it these days. So no, all major models suffer from slop in their training data.
bigthymer 7mo ago
> Can the source database for an LLM be one-way, in that it does not contain output from itself, or other LLMs? I think, for public internet data, we can only be reasonably confident for information before the big release of ChatGPT.
nsvd2 7mo ago
Yes, people have likened pre-LLM Internet content to low-background steel. If in the hypothetical future the continual learning problem gets solved, the AI could just learn from the real world instead of publications and retain that data.
nprateem 7mo ago
One reason why Google made that algorithm to watermark AI output

esperent 7mo ago

Presumably companies will still provide manuals.

PacificSpecific 7mo ago
If that was adequate then wouldn't there not be supplementary material? Results vary of course. I have some very wonderful synthesizer manuals.
SiempreViernes 7mo ago
It'll be a single sheet of paper with a QR code that redirects to a canned prompt hosted at whichever LLM server paid the most to the manufacturer for their content.

roxolotl 7mo ago

Yea so I’ve had an issue getting video output after boot on a new AMD R9700 Pro. None of the, albeit free, models from OpenAI/Google/Anthropic have really been helpful. I found the pro drivers myself. They never mentioned them. Thats not to say AI is bad. It’s great in many cases. More that I’m worried about what happens when the repositories of new knowledge get hollowed out. Also my favorite response was this gem from Sonnet: > TL;DR: Move your monitor cable from the motherboard to the graphics card.

elictronic 7mo ago

This then becomes the hardware manufacturers problem. If their new hardware fails for to many users it will no longer be purchased. If they externalize their problem solving like so many companies, they won't be able to gain market share. This creates financial incentives to pay companies running the new version of search. Your thinking of this as a problem for these companies, when in reality it is a financial incentive.

visarga 7mo ago

> because it has nothing to pull from? Chat rooms produce trillions of tokens per day now, interactive tokens, where AI can poke and prod at us, and have its ideas tested in the real world (by us).