4 ms·

Those answers suck though. I would not consider a human who said them to be intelligent. For the first one, it’s not that you can’t see, it’s specifically that

Those answers suck though. I would not consider a human who said them to be intelligent. For the first one, it’s not that you can’t see, it’s specifically that you will fall and crack your head. An answer that doesn’t mention falling is a bad answer. Number two, it goes off on temperature and sample sizes which is not germane to the question. The answer should tell a story about causality and instead it goes off on ephemera. Three is a Wikipedia search, ie Siri can answer this today, and it’s wrong.

somebodythere 5y ago
All that's missing is a reranking step that rewards incisive and non-obvious answers. Actually, some models are already implementing such a thing: LAmDA will prioritize contentful answers above statistical probability, the same way humans do. GPT was only trained on "text" with no specific distribution, and specifically to predict a masked word from its surrounding context. As a result, the only questions you can use GPT to answer are of the form "sample from the most likely continuations of this text". It turns out a lot of problems can actually be posed in this form, if you understand the input distribution well. But the future probably looks less like models that operate directly on a distribution of language itself, and more like models that use their language knowledge to predict the volition of the person who gave it input, relate the input to their knowledge of the world (learned from corpus), and then use their language knowledge to convert an internal abstract representation of logical reasoning to something the user can read and understand. I don't think the tech is extremely far off, it's probably a natural continuation of the current research.
throwawayboise 5y ago
For some reason I am having a very hard time resisting the urge to try walking down a flight of stairs backwards with my eyes closed. How hard can it be?
- jacquesm 5y ago
  I do that every time I go down a ladder.

somebodythere 5y ago

All that's missing is a reranking step that rewards incisive and non-obvious answers. Actually, some models are already implementing such a thing: LAmDA will prioritize contentful answers above statistical probability, the same way humans do. GPT was only trained on "text" with no specific distribution, and specifically to predict a masked word from its surrounding context. As a result, the only questions you can use GPT to answer are of the form "sample from the most likely continuations of this text". It turns out a lot of problems can actually be posed in this form, if you understand the input distribution well. But the future probably looks less like models that operate directly on a distribution of language itself, and more like models that use their language knowledge to predict the volition of the person who gave it input, relate the input to their knowledge of the world (learned from corpus), and then use their language knowledge to convert an internal abstract representation of logical reasoning to something the user can read and understand. I don't think the tech is extremely far off, it's probably a natural continuation of the current research.

throwawayboise 5y ago

For some reason I am having a very hard time resisting the urge to try walking down a flight of stairs backwards with my eyes closed. How hard can it be?

jacquesm 5y ago
I do that every time I go down a ladder.