3 ms·

Big pretrained models are good enough now that we can pipe them together in really cool ways and our representations of text and images seem to capture what we

Big pretrained models are good enough now that we can pipe them together in really cool ways and our representations of text and images seem to capture what we “mean.”

tpmx 4y ago
Yeah, it seems like it. But it's still just complicated statistical models. Again, where is the reasoning?
- 6gvONxR4sf7o 4y ago
  I don’t care whether it reasons its way from “3 teddy bears below 7 flamingos” to a picture of that or if it gets there some other way. But also, some of the magic in having good enough pretrained representations is that you don’t need to train them further for downstream tasks, which means non-differentiable tasks like logic could soon become more tenable.
- renewiltord 4y ago
  A belief oft shared is that sufficiently complicated statistical models are indistinguishable from reasoning.
- marvin 4y ago
  I still think we're missing some fundamental insights on how layered planning/forecasting/deducting/reasoning works, and that figuring this out will be necessary in order to create AI that we could say "reasons". But with the recent advances/demonstrations, it seems more likely today than in 2019 that our current computational resources are sufficient to perform magnificantly spooky stuff if they're used correctly. They are doing that already already, and that's without deliberately making the software do anything except draw from a vast pool of examples. I think it's reasonable, based on this, to update one's expectations of what we'd be able to do if we figured out ways of doing things that aren't based on first seeing a hundred million examples of what we want the computer to do. Things that do this can obviously exist, we are living examples. Does figuring it out seem likely to be many decades away?
  - tpmx 4y ago
    That's a well-balanced response that I can agree with. I'm an not AGI-skeptic. I'm just a bit skeptical that the topic of this thread is the path forward. It seems to me like an exotic detour. And, of course intelligence isn't magic. We're producing new intelligent entities at rate of a about ~5 per second globally, every day. > Does figuring it out seem likely to be many decades away? 1-7?
- londons_explore 4y ago
  All it takes is one 'trick' to give these models the ability to do reasoning. Like for example the discovery that language models get far better at answering complex questions if asked to show their working step by step with chain of thought reasoning as in page 19 of the PaLM paper [1]. Worth checking out the explanations of novel jokes on page 38 of the same paper. While it is, like you say, all statistics, if it's indistinguishable from valid reasoning, then perhaps it doesn't matter. [1]: https://arxiv.org/pdf/2204.02311.pdf https://arxiv.org/pdf/2204.02311.pdf

Big pretrained models are good enough now that we can pipe them together in really cool ways and our representations of text and images seem to capture what we “mean.”

tpmx 4y ago

Yeah, it seems like it. But it's still just complicated statistical models. Again, where is the reasoning?

6gvONxR4sf7o 4y ago
I don’t care whether it reasons its way from “3 teddy bears below 7 flamingos” to a picture of that or if it gets there some other way. But also, some of the magic in having good enough pretrained representations is that you don’t need to train them further for downstream tasks, which means non-differentiable tasks like logic could soon become more tenable.
renewiltord 4y ago
A belief oft shared is that sufficiently complicated statistical models are indistinguishable from reasoning.
marvin 4y ago
I still think we're missing some fundamental insights on how layered planning/forecasting/deducting/reasoning works, and that figuring this out will be necessary in order to create AI that we could say "reasons". But with the recent advances/demonstrations, it seems more likely today than in 2019 that our current computational resources are sufficient to perform magnificantly spooky stuff if they're used correctly. They are doing that already already, and that's without deliberately making the software do anything except draw from a vast pool of examples. I think it's reasonable, based on this, to update one's expectations of what we'd be able to do if we figured out ways of doing things that aren't based on first seeing a hundred million examples of what we want the computer to do. Things that do this can obviously exist, we are living examples. Does figuring it out seem likely to be many decades away?
- tpmx 4y ago
  That's a well-balanced response that I can agree with. I'm an not AGI-skeptic. I'm just a bit skeptical that the topic of this thread is the path forward. It seems to me like an exotic detour. And, of course intelligence isn't magic. We're producing new intelligent entities at rate of a about ~5 per second globally, every day. > Does figuring it out seem likely to be many decades away? 1-7?
londons_explore 4y ago
All it takes is one 'trick' to give these models the ability to do reasoning. Like for example the discovery that language models get far better at answering complex questions if asked to show their working step by step with chain of thought reasoning as in page 19 of the PaLM paper [1]. Worth checking out the explanations of novel jokes on page 38 of the same paper. While it is, like you say, all statistics, if it's indistinguishable from valid reasoning, then perhaps it doesn't matter. [1]: https://arxiv.org/pdf/2204.02311.pdf https://arxiv.org/pdf/2204.02311.pdf