3 ms·

On one hand, that may be true. On the other hand, we may now be able to provide LLM-powered fact checkers, able to reply to every post or label them in order to

On one hand, that may be true. On the other hand, we may now be able to provide LLM-powered fact checkers, able to reply to every post or label them in order to slow down the dissemination of that kind of info. Of course, authoritarian regimes will be able to do so as well, which is both a curse (much more reactive censorship) and a blessing (unlikely to understand new euphemisms or coded messages by itself).

yowzadave 4y ago
Yes—it seems likely that the effect will be to concentrate power/wealth in the hands of those who already have it, since they will have greater access to the more powerful AIs. In the early days of the internet, it was easy to imagine many positive outcomes for the technology (in addition to the negatives). So far with AI, it seems like the potential negatives are much more salient in people’s imagination.
- musicale 4y ago
  There was a time when PCs seemed to democratize computing away from centralized computing services. But the centralized services eventually came back in new forms. Maybe Personal AI systems will change things up a bit eventually.
CatWChainsaw 4y ago
And how exactly are we going to 100% remove hallucinations from the LLM powered fact checkers?
- MayeulC 3y ago
  Who checks the fact-checkers? More LLMs (semi-seriously). I think there is a pretty clear trend of not leveraging "trained"/"built-in" knowledge if you value accuracy, but rather use LLMs as a way to extract meaning and compare to a known-good reference. One way would be to use LLMs to flag suspicious talking points, compile a list of propaganda topics and their rebuttal, and give that in the context window of LLMs that formulate a response. For more general fact-checking, I guess one could ask LLMs to extract the various points, run a quick web check on them (maybe we need a "facts" database?), and examine the results. This wouldn't protect against logical fallacies though, I'm not sure that can be avoided short of injecting a proof assistant somehow (or providing rebuttals to known talking points, as suggested above).
  - CatWChainsaw 3y ago
    Great, a turtle ouroboros.

yowzadave 4y ago

Yes—it seems likely that the effect will be to concentrate power/wealth in the hands of those who already have it, since they will have greater access to the more powerful AIs. In the early days of the internet, it was easy to imagine many positive outcomes for the technology (in addition to the negatives). So far with AI, it seems like the potential negatives are much more salient in people’s imagination.

musicale 4y ago
There was a time when PCs seemed to democratize computing away from centralized computing services. But the centralized services eventually came back in new forms. Maybe Personal AI systems will change things up a bit eventually.

CatWChainsaw 4y ago

And how exactly are we going to 100% remove hallucinations from the LLM powered fact checkers?

MayeulC 3y ago
Who checks the fact-checkers? More LLMs (semi-seriously). I think there is a pretty clear trend of not leveraging "trained"/"built-in" knowledge if you value accuracy, but rather use LLMs as a way to extract meaning and compare to a known-good reference. One way would be to use LLMs to flag suspicious talking points, compile a list of propaganda topics and their rebuttal, and give that in the context window of LLMs that formulate a response. For more general fact-checking, I guess one could ask LLMs to extract the various points, run a quick web check on them (maybe we need a "facts" database?), and examine the results. This wouldn't protect against logical fallacies though, I'm not sure that can be avoided short of injecting a proof assistant somehow (or providing rebuttals to known talking points, as suggested above).
- CatWChainsaw 3y ago
  Great, a turtle ouroboros.