4 ms·

My mum is a shrink, and very old, and smart, and hates technology. I was talking her through how some primitive "AGI" could happen with 4o (basically just expla

by neom 2y ago

My mum is a shrink, and very old, and smart, and hates technology. I was talking her through how some primitive "AGI" could happen with 4o (basically just explained this: https://b.h4x.zip/agi https://b.h4x.zip/agi) That got us talking about consciousness, and at the end, she thought about it for about a minute and then said "if I can't give it lysergic acid and make it see god, it's not conscious" and went back to making her dinner.

alchemist1e9 2y ago
> She thought about it for about a minute and then said "if I can't give it lysergic acid and make it see god, it's not conscious" and went back to making her dinner. That’s probably going to be pretty easy, just scramble some attention heads randomly in portions of the transformer network and the AGI will probably think “it sees god”.
Zambyte 2y ago
You can give it a virtual device as the real time camera input to show it things that aren't there. That's not organic acid, but the target isn't organic consciousness.
- fuzzfactor 2y ago
  One uncanny thing is that people have such a diverse tolerance for "fake", across a spectrum from absolute acceptance to complete rejection. Or for things like hallucinatory perceptions of reality.
aeonik 2y ago
A psychologist I know did their PhD in this area (also old and smart), and he called this kind of thinking "Neural Chauvanism" -> if her explanation actually requires the chemical and neural components. PDF warning: https://gwern.net/doc/philosophy/mind/1985-cuda.pdf https://gwern.net/doc/philosophy/mind/1985-cuda.pdf
- neom 2y ago
  It's a good point as i'm sure what she was saying is she belives consciousness requires specific neural properties. I am aware of the the ideas around neural chauvanists, but that paper is now almost 40 years old. We know a lot more about the brain since it was written. The idea that the homunculi can perfectly replicate the relevant causal powers of neurons is... questionable at best. It would says distinctive biochemical properties of neurons enable consciousness, which homunculi would lack, and then homunculized, I also don't like the idea that we must attribute consciousness to the homunculized brain unless we accept an implausible cut-off point? Consciousness could be requiring a critical mass of neurons, which could be reached later without each replacement causing incremental fading. The implausibility of a single neuron replacement eliminating consciousness is not a good reason to consider it conscious. imo It's probably still more likley that experiences like an LSD trip require specifically neural underpinnings.
- schmidt_fifty 2y ago
  [dead]
stvltvs 2y ago
That's begging the question of whether inorganic matter can be conscious. If you boil it down, she's just said if it's inorganic, it's not conscious.
- neom 2y ago
  It's funny, I've studied a lot of psychedelics (probably because it was an area of research for both my parents), and salvia divinorum is a really really stand out plant in it's trips, it seems to be a very "technically philosophical" plant. Trip reports always go into weird things like "I became a book on a shelf for 4,000 years, and now I know inanimate objects are conscious", there is also the area of panpsychism and animism. That stuff is all a little too mind bending for me, but pretty fun thinking for a Sunday morning. :)
  - eaenki 2y ago
    well, i tried salvia and I tried shrooms. I don’t see how it’s more philosophical just because it’s a dissociative. at best, it’s more of an ego death than, say, shrooms, if during the trip you become an object. It’s still a unique plant for sure because of it’s dissociative effects. To thread lightly. AFAIK natives in mexico say that smoking it is very bad. and that u’re supposed to chew fresh leaves. maybe chewing fresh leaves doesn’t even cause the same dissociative effects which are unpleasant.
mgdante 2y ago
Raising the temperature is probably a close analog to psychedelics.
- boesboes 2y ago
  I've read once they determined the mechanism by which a.o lsd 'works', is by lowering/disabling a lot of the filtering between neurons. This leads us to recognise all kinds of patterns that are not really there. Visual hallucinations being the obvious form of this, but I suppose the same applies to other things like our personalities and self; that's neither here nor there anyway. Now I'm not 100% sure how temperature is implemented, but from what i recollect, might be a reasonable analogue indeed!
  - neom 2y ago
    That's not really correct. default mode network activity is somewhat disrupted, but LSD works primarily by interaction with the 5-HT2A receptor subtype, this is multiple neurotransmitter systems, including serotonin, dopamine, and glutamate, and in many examples it's via excitement not inhibition/lowering/disabling. Excitatory neurotransmission in glutamate is looking like the the most important of the work in the 5-HT2A area.
    - schmidtleonard 2y ago
      Does excitatory/inhibitory map cleanly on to the higher level abstractions? I'm no neuroscientist, so maybe it does, but that seems like a bad assumption to me because in digital logic "active when low" is extremely common. A spurious-suppression system that caused inhibition when low would be perfectly reasonable and compatible with the observation that excitation caused an increase in spurious behavior.
      - neom 2y ago
        They don't really map well. The brain obviously operates on more continuous principles rather than binary states. The relationship between excitation/inhibition and emergent effects may be more direct and graded in biological neural networks vs to the "active when low". Buuuutt stiilll, 5-HT2A receptor and its downstream effects on glutamate transmission play a central role in mediating the subjective effects of psychedelics. Very very many studies have consistently linked 5-HT2A activation to the perceptual, cognitive, and emotional characteristic of the psychedelic state. Disruption of the default mode network and other changes are somewhat important maybe, the 5-HT2A receptor appears to be the focus of action for producing these effects. The point being, neurochemical interactions at the molecular level are the most important aspects of the LSD interaction, so.., now sure how that works in the context of synthetic AI, I don't know much about ML/NN.
killerstorm 2y ago
Well, this guy is highly proficient at administering an artificial lysergic acid to LLMs: https://x.com/repligate/status/1792010019744960577 https://x.com/repligate/status/1792010019744960577
- schmidtleonard 2y ago
  Deep Dream is wild: https://www.youtube.com/watch?v=5DaVnriHhPc https://www.youtube.com/watch?v=5DaVnriHhPc
- moffkalast 2y ago
  And a few drinks too many: https://www.reddit.com/r/LocalLLaMA/comments/13vv941/temperature10000_and_top_p01_is_hilarious/ https://www.reddit.com/r/LocalLLaMA/comments/13vv941/tempera...
bee_rider 2y ago
That seems more like a good natured refusal to engage with the question seriously. Sort of like, if you come to a smart engineer with a design for a perpetual motion machine, they might likely tease you a little bit and then refer you to a physicist. Smart people from applied fields know when the topics are outside their actual wheelhouse, but getting close enough that they risk being taken seriously, to a misleading extent.
- hn_throwaway_99 2y ago
  As someone whose experience with ketamine therapy profoundly changed my life, I think she was making a slightly different, sly point. As someone who isn't religious, and who doesn't have children, and who hit middle age a bit disillusioned with my career, I struggled a lot with my life's purpose (nevermind chronic depression). But I don't think it's really an exaggeration to say that a single dose of ketamine therapy (this was an hour long infusion at a clinic) made me understand that for me, really, consciousness and existence is enough of a reason for my life's purpose. Throughout nearly my entire trip I kept thinking "how amazing is it that my brain doing this???" I think I understand computational theories of the mind well enough, but they in no way explain why consciousness would "fall out" of computation. I think people sort of intuitively grasp that consciousness is something really different than the "normal" physical world. I think the quote of "if I can't give it lysergic acid and make it see god, it's not conscious" is about the fact that psychedelics provide even more clear insight into the "consciousness is something different" idea. Some people interpret that as God. As a non-religious person I didn't necessarily interpret that as "God", but I did interpret it as consciousness being something wholly different than normal physical matter or just the "math of linear algebra". For me it's a wondrous, magical thing, and I'm just glad I get to experience it.
agumonkey 2y ago
Funny story, but at the same time, got seems fine about hallucinating by itself. At least a little.
Der_Einzige 2y ago
You can give LLMs DRuGS though! https://github.com/EGjoni/DRUGS https://github.com/EGjoni/DRUGS
m463 2y ago
Can't you train a model on LSD experiences, and use that as a prompt? https://www.openculture.com/2017/08/artist-draws-a-series-portraits-on-lsd-inside-the-1950s-experiments-to-turn-lsd-into-a-creativity-pill.html https://www.openculture.com/2017/08/artist-draws-a-series-po... or some sort of reverse stable diffusion training? https://duckduckgo.com/?q=lsd+trip+artist&iax=images&ia=images https://duckduckgo.com/?q=lsd+trip+artist&iax=images&ia=imag...

by neom 2y ago

alchemist1e9 2y ago

> She thought about it for about a minute and then said "if I can't give it lysergic acid and make it see god, it's not conscious" and went back to making her dinner. That’s probably going to be pretty easy, just scramble some attention heads randomly in portions of the transformer network and the AGI will probably think “it sees god”.

Zambyte 2y ago

You can give it a virtual device as the real time camera input to show it things that aren't there. That's not organic acid, but the target isn't organic consciousness.

fuzzfactor 2y ago
One uncanny thing is that people have such a diverse tolerance for "fake", across a spectrum from absolute acceptance to complete rejection. Or for things like hallucinatory perceptions of reality.

aeonik 2y ago

A psychologist I know did their PhD in this area (also old and smart), and he called this kind of thinking "Neural Chauvanism" -> if her explanation actually requires the chemical and neural components. PDF warning: https://gwern.net/doc/philosophy/mind/1985-cuda.pdf https://gwern.net/doc/philosophy/mind/1985-cuda.pdf

neom 2y ago
It's a good point as i'm sure what she was saying is she belives consciousness requires specific neural properties. I am aware of the the ideas around neural chauvanists, but that paper is now almost 40 years old. We know a lot more about the brain since it was written. The idea that the homunculi can perfectly replicate the relevant causal powers of neurons is... questionable at best. It would says distinctive biochemical properties of neurons enable consciousness, which homunculi would lack, and then homunculized, I also don't like the idea that we must attribute consciousness to the homunculized brain unless we accept an implausible cut-off point? Consciousness could be requiring a critical mass of neurons, which could be reached later without each replacement causing incremental fading. The implausibility of a single neuron replacement eliminating consciousness is not a good reason to consider it conscious. imo It's probably still more likley that experiences like an LSD trip require specifically neural underpinnings.
schmidt_fifty 2y ago
[dead]

stvltvs 2y ago

That's begging the question of whether inorganic matter can be conscious. If you boil it down, she's just said if it's inorganic, it's not conscious.

neom 2y ago
It's funny, I've studied a lot of psychedelics (probably because it was an area of research for both my parents), and salvia divinorum is a really really stand out plant in it's trips, it seems to be a very "technically philosophical" plant. Trip reports always go into weird things like "I became a book on a shelf for 4,000 years, and now I know inanimate objects are conscious", there is also the area of panpsychism and animism. That stuff is all a little too mind bending for me, but pretty fun thinking for a Sunday morning. :)
- eaenki 2y ago
  well, i tried salvia and I tried shrooms. I don’t see how it’s more philosophical just because it’s a dissociative. at best, it’s more of an ego death than, say, shrooms, if during the trip you become an object. It’s still a unique plant for sure because of it’s dissociative effects. To thread lightly. AFAIK natives in mexico say that smoking it is very bad. and that u’re supposed to chew fresh leaves. maybe chewing fresh leaves doesn’t even cause the same dissociative effects which are unpleasant.

mgdante 2y ago

Raising the temperature is probably a close analog to psychedelics.

boesboes 2y ago
I've read once they determined the mechanism by which a.o lsd 'works', is by lowering/disabling a lot of the filtering between neurons. This leads us to recognise all kinds of patterns that are not really there. Visual hallucinations being the obvious form of this, but I suppose the same applies to other things like our personalities and self; that's neither here nor there anyway. Now I'm not 100% sure how temperature is implemented, but from what i recollect, might be a reasonable analogue indeed!
- neom 2y ago
  That's not really correct. default mode network activity is somewhat disrupted, but LSD works primarily by interaction with the 5-HT2A receptor subtype, this is multiple neurotransmitter systems, including serotonin, dopamine, and glutamate, and in many examples it's via excitement not inhibition/lowering/disabling. Excitatory neurotransmission in glutamate is looking like the the most important of the work in the 5-HT2A area.
  - schmidtleonard 2y ago
    Does excitatory/inhibitory map cleanly on to the higher level abstractions? I'm no neuroscientist, so maybe it does, but that seems like a bad assumption to me because in digital logic "active when low" is extremely common. A spurious-suppression system that caused inhibition when low would be perfectly reasonable and compatible with the observation that excitation caused an increase in spurious behavior.
    - neom 2y ago
      They don't really map well. The brain obviously operates on more continuous principles rather than binary states. The relationship between excitation/inhibition and emergent effects may be more direct and graded in biological neural networks vs to the "active when low". Buuuutt stiilll, 5-HT2A receptor and its downstream effects on glutamate transmission play a central role in mediating the subjective effects of psychedelics. Very very many studies have consistently linked 5-HT2A activation to the perceptual, cognitive, and emotional characteristic of the psychedelic state. Disruption of the default mode network and other changes are somewhat important maybe, the 5-HT2A receptor appears to be the focus of action for producing these effects. The point being, neurochemical interactions at the molecular level are the most important aspects of the LSD interaction, so.., now sure how that works in the context of synthetic AI, I don't know much about ML/NN.

killerstorm 2y ago

Well, this guy is highly proficient at administering an artificial lysergic acid to LLMs: https://x.com/repligate/status/1792010019744960577 https://x.com/repligate/status/1792010019744960577

schmidtleonard 2y ago
Deep Dream is wild: https://www.youtube.com/watch?v=5DaVnriHhPc https://www.youtube.com/watch?v=5DaVnriHhPc
moffkalast 2y ago
And a few drinks too many: https://www.reddit.com/r/LocalLLaMA/comments/13vv941/temperature10000_and_top_p01_is_hilarious/ https://www.reddit.com/r/LocalLLaMA/comments/13vv941/tempera...

bee_rider 2y ago

That seems more like a good natured refusal to engage with the question seriously. Sort of like, if you come to a smart engineer with a design for a perpetual motion machine, they might likely tease you a little bit and then refer you to a physicist. Smart people from applied fields know when the topics are outside their actual wheelhouse, but getting close enough that they risk being taken seriously, to a misleading extent.

hn_throwaway_99 2y ago
As someone whose experience with ketamine therapy profoundly changed my life, I think she was making a slightly different, sly point. As someone who isn't religious, and who doesn't have children, and who hit middle age a bit disillusioned with my career, I struggled a lot with my life's purpose (nevermind chronic depression). But I don't think it's really an exaggeration to say that a single dose of ketamine therapy (this was an hour long infusion at a clinic) made me understand that for me, really, consciousness and existence is enough of a reason for my life's purpose. Throughout nearly my entire trip I kept thinking "how amazing is it that my brain doing this???" I think I understand computational theories of the mind well enough, but they in no way explain why consciousness would "fall out" of computation. I think people sort of intuitively grasp that consciousness is something really different than the "normal" physical world. I think the quote of "if I can't give it lysergic acid and make it see god, it's not conscious" is about the fact that psychedelics provide even more clear insight into the "consciousness is something different" idea. Some people interpret that as God. As a non-religious person I didn't necessarily interpret that as "God", but I did interpret it as consciousness being something wholly different than normal physical matter or just the "math of linear algebra". For me it's a wondrous, magical thing, and I'm just glad I get to experience it.

agumonkey 2y ago

Funny story, but at the same time, got seems fine about hallucinating by itself. At least a little.

Der_Einzige 2y ago

You can give LLMs DRuGS though! https://github.com/EGjoni/DRUGS https://github.com/EGjoni/DRUGS

m463 2y ago

Can't you train a model on LSD experiences, and use that as a prompt? https://www.openculture.com/2017/08/artist-draws-a-series-portraits-on-lsd-inside-the-1950s-experiments-to-turn-lsd-into-a-creativity-pill.html https://www.openculture.com/2017/08/artist-draws-a-series-po... or some sort of reverse stable diffusion training? https://duckduckgo.com/?q=lsd+trip+artist&iax=images&ia=images https://duckduckgo.com/?q=lsd+trip+artist&iax=images&ia=imag...