3 ms·

I think what is also interesting is what LLMs help us learn about ourselves. I think we understand human intelligence a bit better because of LLMs, and I think

I think what is also interesting is what LLMs help us learn about ourselves. I think we understand human intelligence a bit better because of LLMs, and I think people have been surprised how far you could get just by processing text. I think (just a guess relying on intuition and my very shallow survey) that human intelligence is multilayered where we have something at least vaguely analogous to a LLM combined with something else that knows the physics of the world as well as something else that is able to do logic and symbolic processing. But this is exciting to see the power of relatively simple neural networks when they are trained with a giant corpus.

MeImCounting 3y ago
I really believe this to be the case. Its remarkable what language alone can do but its not the be all end all of intelligence.
bormaj 3y ago
Did the prevalence of LLMs shed any light on what intelligence is vs what it is not?
- paganel 3y ago
  Not that I personally know of, but it did create (or empowered, because in a way it was already there) a new sort of anthropomorphization, of which this article is partially guilty of itself (the part labeled "LLMs are really smart, and also really, really dumb"). Some of us now think they know more about human intelligence as a result of that anthropomorphization.
growingkittens 3y ago
I find something lacking from many AI discussions because my experience with having a brain is radically different from average. Mine is damaged by a childhood brain injury, which wasn't diagnosed until I was 30. Another thing about the injury is that my lower order thinking skills seem to be damaged, such as sequencing or anything based on working memory. However, my higher order thinking skills are intact - I think they've been compensating for the broken skills this whole time, albeit with much difficulty. My thoughts on LLMs from a previous comment: > I think in pictures; I remember information with emotions, music, movement, and basically anywhere my brain could stuff information. Words are ephemeral, I often forget the beginning of a sentence by the time I reach the end. > The human brain is so much more than language. Also: > There are many specialists and not many generalists. Generalists are needed to tie together the multiple domains required to emulate the human brain.
- Terretta 3y ago
  By contrast, particularly observing a range of adults from colleagues to seniors with early onset dementia, I'm increasingly sure a large portion of discourse is of the "LLM continuation" variety. Everyone knows that colleague (I am one myself) where a particular topic will kick off a continuation that follows a well worn path albeit with a bit higher temperature than zero. With early onset dementia where working memory and self-suppression are removed, you can unspool the same commentary as often as you like by teeing up the same context to trigger it. See also these layers, nothing to do with LLM -- Attention, Language, Memory, Executive Function: https://academic.oup.com/view-large/figure/198261184/awz311f1.png https://academic.oup.com/view-large/figure/198261184/awz311f... Or maybe quite a lot to do with it.
  - growingkittens 3y ago
    > With early onset dementia where working memory and self-suppression are removed, you can unspool the same commentary as often as you like by teeing up the same context to trigger it. My working memory is pretty much shot, between inherited ADHD and a brain injury. Whatever is in my field of vision makes up the majority of my reality - associations with things I see, and the trains of thought that expand from those associations. The rest of my reality comes from whatever memories my brain conjures up when it has nothing to focus on. I use what I call "external prompts" to function on a daily basis, or 'teeing up the same context' to trigger a memory to trigger an action. Something that concealed my brain injury as a child was how well I did for most parts of school. I did well because tests are a series of external prompts, and my long term memory is fairly accurate. My math memory is terrible, though, as I can only hold 1-2 numbers in my mind at a time. For instance, to remember the time, I always round to the nearest 5 or 0. 2:43 becomes 2:45. It takes less number memory when the last option is only a 5 or 0. I also use picture memory to temporarily store the exact time (as a snapshot from when I looked at the clock), although the numbers start mixing with previous snapshots.
- layer8 3y ago
  Your quote doesn’t sound very different from what undamaged brains do (based on my personal n = 1 experience).
rorroe53 3y ago
Human intelligence is definitely multilayered, and different parts of brain are known to be responsible for different types of processing. Personally I was born with a disability that makes me suck at motor skills, spatial learning and understanding most forms of mathematics, but somehow I'm still decent at programming, and significantly better than average at reading and writing - all of which LLMs can do quite well compared to its mathematics and logic skills.
archagon 3y ago
I’ve noticed myself looking at art recently with LLMs in mind. Is this something that an AI could spit out, or is there evidence of a human touch? And when I create art, how do I ensure it doesn’t end up looking like something out of Midjourney? I think we will learn a lot about the creative process from AI, and I think human-created art will get a lot more interesting (and weird) as a result.

MeImCounting 3y ago

I really believe this to be the case. Its remarkable what language alone can do but its not the be all end all of intelligence.

bormaj 3y ago

Did the prevalence of LLMs shed any light on what intelligence is vs what it is not?

paganel 3y ago
Not that I personally know of, but it did create (or empowered, because in a way it was already there) a new sort of anthropomorphization, of which this article is partially guilty of itself (the part labeled "LLMs are really smart, and also really, really dumb"). Some of us now think they know more about human intelligence as a result of that anthropomorphization.

growingkittens 3y ago

I find something lacking from many AI discussions because my experience with having a brain is radically different from average. Mine is damaged by a childhood brain injury, which wasn't diagnosed until I was 30. Another thing about the injury is that my lower order thinking skills seem to be damaged, such as sequencing or anything based on working memory. However, my higher order thinking skills are intact - I think they've been compensating for the broken skills this whole time, albeit with much difficulty. My thoughts on LLMs from a previous comment: > I think in pictures; I remember information with emotions, music, movement, and basically anywhere my brain could stuff information. Words are ephemeral, I often forget the beginning of a sentence by the time I reach the end. > The human brain is so much more than language. Also: > There are many specialists and not many generalists. Generalists are needed to tie together the multiple domains required to emulate the human brain.

Terretta 3y ago
By contrast, particularly observing a range of adults from colleagues to seniors with early onset dementia, I'm increasingly sure a large portion of discourse is of the "LLM continuation" variety. Everyone knows that colleague (I am one myself) where a particular topic will kick off a continuation that follows a well worn path albeit with a bit higher temperature than zero. With early onset dementia where working memory and self-suppression are removed, you can unspool the same commentary as often as you like by teeing up the same context to trigger it. See also these layers, nothing to do with LLM -- Attention, Language, Memory, Executive Function: https://academic.oup.com/view-large/figure/198261184/awz311f1.png https://academic.oup.com/view-large/figure/198261184/awz311f... Or maybe quite a lot to do with it.
- growingkittens 3y ago
  > With early onset dementia where working memory and self-suppression are removed, you can unspool the same commentary as often as you like by teeing up the same context to trigger it. My working memory is pretty much shot, between inherited ADHD and a brain injury. Whatever is in my field of vision makes up the majority of my reality - associations with things I see, and the trains of thought that expand from those associations. The rest of my reality comes from whatever memories my brain conjures up when it has nothing to focus on. I use what I call "external prompts" to function on a daily basis, or 'teeing up the same context' to trigger a memory to trigger an action. Something that concealed my brain injury as a child was how well I did for most parts of school. I did well because tests are a series of external prompts, and my long term memory is fairly accurate. My math memory is terrible, though, as I can only hold 1-2 numbers in my mind at a time. For instance, to remember the time, I always round to the nearest 5 or 0. 2:43 becomes 2:45. It takes less number memory when the last option is only a 5 or 0. I also use picture memory to temporarily store the exact time (as a snapshot from when I looked at the clock), although the numbers start mixing with previous snapshots.
layer8 3y ago
Your quote doesn’t sound very different from what undamaged brains do (based on my personal n = 1 experience).

rorroe53 3y ago

Human intelligence is definitely multilayered, and different parts of brain are known to be responsible for different types of processing. Personally I was born with a disability that makes me suck at motor skills, spatial learning and understanding most forms of mathematics, but somehow I'm still decent at programming, and significantly better than average at reading and writing - all of which LLMs can do quite well compared to its mathematics and logic skills.

archagon 3y ago

I’ve noticed myself looking at art recently with LLMs in mind. Is this something that an AI could spit out, or is there evidence of a human touch? And when I create art, how do I ensure it doesn’t end up looking like something out of Midjourney? I think we will learn a lot about the creative process from AI, and I think human-created art will get a lot more interesting (and weird) as a result.