3 ms·

> limitations of probabilistic models that hallucinate, where they authoritatively state things that aren't true I really don't understand why people think it'

> limitations of probabilistic models that hallucinate, where they authoritatively state things that aren't true I really don't understand why people think it's some insurmountable wall. People do hallucinate in the same way llm-s do. If you talk to a baby that just acquired language it spews all sorts of nonsense, but also some sensible things. So far we made a baby, granted, with huge vocabulary and shallow knowledge of many things. But now we need to raise it. The same way we raise humans. With better training. By exposing it to new knowlege but also reducing input of garbage. By making it hear the words that have more and more complex concepts behind them and telling it when it generates outputs that are erroneous. It's a monumental task, but we are not raising an only child. We need to figure out how to make children cooperate and teach from each other. Synthetic data will be a huge part of this process. Gold and shovels ... AI might be search for gold, but creating variety of synthetic data from deep reasoning will be shovels factory. It's weird that people suddenly expect 4 year old to accurately reproduce encyclopedia just because it can talk now.

dartos 2y ago
> I really don't understand why people think it's some insurmountable wall. People do hallucinate in the same way llm-s do. Go talk to some real people. A person making a mistake is not similar to a statistical model’s prediction not matching reality. It’s weird that you think probabilistic a text generator is even nearly comparable to an organic brain (let alone a human) It’s like comparing a wax statue with flesh and blood. They may look similar, they last a lot longer, but they’re not comparable. You’ve fallen totally into personification bias. I hope you find your way out.
- snapcaster 2y ago
  Nothing in your comment argues against it being surmountable
  - dartos 2y ago
    Nothing in my comment argues against the existence of bees either. I have some opinions on the matter, but arguments about unfalsifiable traits aren’t really useful, especially since you’re not going to be convinced, nor I. I don’t personally believe that we’ll get much further with regards to general intelligence with statistical models, specifically. I think a lot of the purported progress towards “intelligence” is personification bias and a little bit of p-hacking. Many of the papers that are published and spread aren’t peer-reviewed or all that reproducible. We’ll probably see some very useful models made soon which focus on narrower text generation tasks and achieve pretty good results.
    - snapcaster 2y ago
      >We’ll probably see some very useful models made soon which focus on narrower text generation tasks and achieve pretty good results. isn't this surmounting the problem? Also aren't there a whole range of techniques already developed to handle hallucinations? I'm not sure what you think is the showstopper here?
      - dartos 2y ago
        > isn't this surmounting the problem? I’m sorry, I’m not really sure what you mean by this. There are techniques to lower the rate of hallucinations, but not solve it. Those techniques also vary wildly from application to application. So, no. There are no general viable techniques to totally avoid hallucinations.
        snapcaster 2y ago
        but totally avoiding isn't necessary? My company is using LLMs in production across a wide range of features right now so i'm not just speaking theoretically edit: including in cases where outputs need to meet strict constraints
monkeydreams 2y ago
> It's weird that people suddenly expect 4 year old to accurately reproduce encyclopedia just because it can talk now. But we are not talking of a four year old but a system wthat has devoured literally all of the works humans have produced to date. If we are going to see further improvement I am keen to see where this emerges from.

dartos 2y ago

> I really don't understand why people think it's some insurmountable wall. People do hallucinate in the same way llm-s do. Go talk to some real people. A person making a mistake is not similar to a statistical model’s prediction not matching reality. It’s weird that you think probabilistic a text generator is even nearly comparable to an organic brain (let alone a human) It’s like comparing a wax statue with flesh and blood. They may look similar, they last a lot longer, but they’re not comparable. You’ve fallen totally into personification bias. I hope you find your way out.

snapcaster 2y ago
Nothing in your comment argues against it being surmountable
- dartos 2y ago
  Nothing in my comment argues against the existence of bees either. I have some opinions on the matter, but arguments about unfalsifiable traits aren’t really useful, especially since you’re not going to be convinced, nor I. I don’t personally believe that we’ll get much further with regards to general intelligence with statistical models, specifically. I think a lot of the purported progress towards “intelligence” is personification bias and a little bit of p-hacking. Many of the papers that are published and spread aren’t peer-reviewed or all that reproducible. We’ll probably see some very useful models made soon which focus on narrower text generation tasks and achieve pretty good results.
  - snapcaster 2y ago
    >We’ll probably see some very useful models made soon which focus on narrower text generation tasks and achieve pretty good results. isn't this surmounting the problem? Also aren't there a whole range of techniques already developed to handle hallucinations? I'm not sure what you think is the showstopper here?
    - dartos 2y ago
      > isn't this surmounting the problem? I’m sorry, I’m not really sure what you mean by this. There are techniques to lower the rate of hallucinations, but not solve it. Those techniques also vary wildly from application to application. So, no. There are no general viable techniques to totally avoid hallucinations.
      - snapcaster 2y ago
        but totally avoiding isn't necessary? My company is using LLMs in production across a wide range of features right now so i'm not just speaking theoretically edit: including in cases where outputs need to meet strict constraints

monkeydreams 2y ago

> It's weird that people suddenly expect 4 year old to accurately reproduce encyclopedia just because it can talk now. But we are not talking of a four year old but a system wthat has devoured literally all of the works humans have produced to date. If we are going to see further improvement I am keen to see where this emerges from.