3 ms·

Maybe in casual conversation but that's not how I experience my though process about anything non trivial at all. I usually spend a lot of time thinking about t

Maybe in casual conversation but that's not how I experience my though process about anything non trivial at all. I usually spend a lot of time thinking about the concept in non verbal terms and that process involves recalling images and sensory information in fairly abstract terms and then through what feels like several iterations it starts to coalesce into something I can encode in language. I think we can all agree that these LLMs are surprisingly good at generating text that is often coherent but I don't see how you can discard all those extra inputs and claim you have the same process.

panarky 3y ago
> thinking about the concept in non verbal ... abstract terms The abstract terms we think about are concepts, and we think about multiple concepts, at various levels of abstraction, and their relationships to each other, before getting a sense of what we want to say or write. Only then do we begin speaking or writing, grouping concepts into paragraphs, breaking them down into sentences and words. And there's evidence that LLMs do something similar, creating embeddings for both big ideas and small details, modeling how the small details combine into larger concepts, discovering the relationships between concepts, and only then generating a probabilistic sequence of tokens to express those deeper concepts.
- cageface 3y ago
  Right but they're limited to concepts that are derived from linguistic tokens. Human reasoning can include nonverbal inputs. For example asking GPT what would be a reasonable way to stack a list of arbitrary objects on top of each other.
  - namaria 3y ago
    Lately I've been think that staple of science fiction, the android robot, is the only way to get AGI. I thought a lot about computer vision when that was the bleeding edge of AI a few years back. It strikes me as obviously wrong to try training models on two dimensional arrays of pixels and expect them to develop visual skills. Why don't we give them two cameras and arms so they can learn about physics and start differentiating 3 dimensional input? I think when we have robots capable of interacting with the world as humans do we might be able to train them to develop near human levels of intelligence.

panarky 3y ago

> thinking about the concept in non verbal ... abstract terms The abstract terms we think about are concepts, and we think about multiple concepts, at various levels of abstraction, and their relationships to each other, before getting a sense of what we want to say or write. Only then do we begin speaking or writing, grouping concepts into paragraphs, breaking them down into sentences and words. And there's evidence that LLMs do something similar, creating embeddings for both big ideas and small details, modeling how the small details combine into larger concepts, discovering the relationships between concepts, and only then generating a probabilistic sequence of tokens to express those deeper concepts.

cageface 3y ago
Right but they're limited to concepts that are derived from linguistic tokens. Human reasoning can include nonverbal inputs. For example asking GPT what would be a reasonable way to stack a list of arbitrary objects on top of each other.
- namaria 3y ago
  Lately I've been think that staple of science fiction, the android robot, is the only way to get AGI. I thought a lot about computer vision when that was the bleeding edge of AI a few years back. It strikes me as obviously wrong to try training models on two dimensional arrays of pixels and expect them to develop visual skills. Why don't we give them two cameras and arms so they can learn about physics and start differentiating 3 dimensional input? I think when we have robots capable of interacting with the world as humans do we might be able to train them to develop near human levels of intelligence.