on the frontier lead dev @hyperware_ai powering irl Jarvis prev @re_meeting @uchicago
Joined January 2024
- Tweets 1,393
- Following 67
- Followers 209
- Likes 2,936
Thinking: methodical, regimented, sequential, linear, slow
Inspiration: sudden appearing of a solution, hypothesis, intuition, etc. Highly nonlinear
Daydreaming: wandering, uncontrolled and uncontrollable, nonlinear, fast
Anyone working hard on difficult problems has experienced the combination of thinking and inspiration, which I’ll call “contemplation “. Thinking about a specific, hard goal will often yield no results. But then, upon taking a break, inspiration strikes. The conscious mind directs and preconditions the unconscious or subconscious, setting the initial conditions for inspiration
I’ll call the combination of thinking and daydreaming “meditation”. I haven’t heard of as many people taking advantage of “meditation”. An anecdote about Einstein comes to mind: he would, so it goes, sit thinking about his problem with a pencil on his knee. He would let himself drift off to sleep: as soon as he fell asleep, the pencil would fall and wake him. He would then write down his insights immediately before they fled his conscious mind
Anyone who has watched their mind while daydreaming or falling asleep will understand how fast daydreaming is compared to thinking. Controlling its flow using “meditation”, similar to “contemplation” controlling inspiration, sounds like a huge unlock. Why don’t more people do it?
“Vibecoding is just managing”
Yes but I don’t have to deal with managing people
No hiring
No firing
Poor performance? Switch LLMs
Didn’t do what I wanted? Restart from scratch
Always make a TODO list
Was important before LLM coding, even more important now: much easier to maintain parallel working agents when you’ve got a TODO list to refer to
It’s cozy season. Grab a warm beverage and read the @Hyperware_ai whitepaper and tokenomics paper
(Link below)
Just watched @pmarca talking about how great Once Upon a Time in Hollywood is. I thought it was one of Tarantino’s worst
Read Echopraxia, the sequel. It’s not as good. If you liked Blindsight it’s worth a read, but just not as good in general: pacing is a bit slow; topics not as compelling
I read Blindsight after carmack recommended it. Enjoyed and would recommend it as well
Hereafter follow musings, including spoilers from the book, so stop reading if you do not want spoilers
Synthesists: rotating information and “patterns have an intelligence all their own”
Very on point prediction, given the surprising effectiveness of LLMs. It seems that, indeed, patterns have an intelligence all their own. I’m still not quite sure all the consequences of this idea
Consciousness is useless (and costly)
Perhaps the main theme of the book was that consciousness is metabolically costly - undeniable - and useless - debatable. Consciousness is certainly metabolically costly: the brain is expensive and any calories spent on building it and powering it could’ve been spent elsewhere (survival, reproduction)
Useless? the argument goes that all the really valuable processing happens at lower levels. There are obvious examples like heart beat: not controlled, or even “conscious” actions like moving your limbs. But even more subtle are examples like creative work. Anyone involved in a creative enterprise has experienced hitting a problem that just couldn’t be solved - until taking a break, or sleeping, and then, suddenly, the solution presented itself. The conscious mind did not solve the problem: the solution bubbled up from below
So what good is the conscious mind? goes the argument. A non-conscious Homo sapiens would be more cost effective and, allegedly, equally effective. A similar (naive) argument can be made that cooperation is doomed, since if you have a cooperating population, a genetic variant that defects gets the benefit of cooperation without the cost, and so in a few generations, everyone will defect
Of course, we do see cooperation, so the naive model is not the whole story. Game theory tells us that iterated prisoners dilemmas are one way out. Genetic relatedness is another. In general, there are many ways out of this trap, and so I suspect there is a similar argument to be made here against “consciousness is useless”
NPCs, genetic and behavioral
As discussed above, consciousness is claimed to be costly and useless. In the book, vampires, a less conscious, more intelligent Homo sapiens variant, end up outcompeting humans back on earth, the same as the naive story told above for cooperation
There isn’t a reason non-consciousness has to be in so starkly different a creature as vampires vs humans, though. This brought to mind “NPCs”. You could explain the claim that some people are NPCs through a similar model: there are literal genetic variants that are pure imitators of consciousness: they don’t pay the cost of consciousness and so, to the extent consciousness gives no advantage, they will take over the entire population as time passes
Less naively you can imagine different people have different predilections to spending time in their heads - almost certainly Gaussian distributed. Depending on your starting conditions you’ll be pushed to introspect, or not, and so will end up with some overall tendency to introspection
This implies some people will end up as NPCs, through behavior and genetics, and others will end up spending all their time in their heads
This isn’t really surprising. Most people’s opinions and beliefs are from others: memetics and mimesis. Almost always the conventional wisdom will be better than what I can cook up myself, and it’s less costly, so why should I do anything other than copy? Even people on the frontier are only on one, or Maybe a few frontiers. It’s extremely rare for someone to be on multiple frontiers: it’s hard enough to be on one!
In closing I’ll say I was surprised that the final two twists of the book were, first, the AI was running the vampire the whole time, and then second, that the vampires took over back home. It seems more plausible to me that if the AI is smarter, and it’s all about smartness sans consciousness, the AIs should’ve taken over
So this is really an AI doomer book in disguise
If the premise is right, the debate about LLM consciousness is irrelevant since it just matters if they are smarter. If you give them agency, the argument would go, it’s game over
Fortunately im highly suspicious of the premise! I think consciousness is pretty good. I’ll post again once I have some specific hypotheses about what specifically it is good for
I read Blindsight after carmack recommended it. Enjoyed and would recommend it as well
Hereafter follow musings, including spoilers from the book, so stop reading if you do not want spoilers
Synthesists: rotating information and “patterns have an intelligence all their own”
Very on point prediction, given the surprising effectiveness of LLMs. It seems that, indeed, patterns have an intelligence all their own. I’m still not quite sure all the consequences of this idea
Consciousness is useless (and costly)
Perhaps the main theme of the book was that consciousness is metabolically costly - undeniable - and useless - debatable. Consciousness is certainly metabolically costly: the brain is expensive and any calories spent on building it and powering it could’ve been spent elsewhere (survival, reproduction)
Useless? the argument goes that all the really valuable processing happens at lower levels. There are obvious examples like heart beat: not controlled, or even “conscious” actions like moving your limbs. But even more subtle are examples like creative work. Anyone involved in a creative enterprise has experienced hitting a problem that just couldn’t be solved - until taking a break, or sleeping, and then, suddenly, the solution presented itself. The conscious mind did not solve the problem: the solution bubbled up from below
So what good is the conscious mind? goes the argument. A non-conscious Homo sapiens would be more cost effective and, allegedly, equally effective. A similar (naive) argument can be made that cooperation is doomed, since if you have a cooperating population, a genetic variant that defects gets the benefit of cooperation without the cost, and so in a few generations, everyone will defect
Of course, we do see cooperation, so the naive model is not the whole story. Game theory tells us that iterated prisoners dilemmas are one way out. Genetic relatedness is another. In general, there are many ways out of this trap, and so I suspect there is a similar argument to be made here against “consciousness is useless”
NPCs, genetic and behavioral
As discussed above, consciousness is claimed to be costly and useless. In the book, vampires, a less conscious, more intelligent Homo sapiens variant, end up outcompeting humans back on earth, the same as the naive story told above for cooperation
There isn’t a reason non-consciousness has to be in so starkly different a creature as vampires vs humans, though. This brought to mind “NPCs”. You could explain the claim that some people are NPCs through a similar model: there are literal genetic variants that are pure imitators of consciousness: they don’t pay the cost of consciousness and so, to the extent consciousness gives no advantage, they will take over the entire population as time passes
Less naively you can imagine different people have different predilections to spending time in their heads - almost certainly Gaussian distributed. Depending on your starting conditions you’ll be pushed to introspect, or not, and so will end up with some overall tendency to introspection
This implies some people will end up as NPCs, through behavior and genetics, and others will end up spending all their time in their heads
This isn’t really surprising. Most people’s opinions and beliefs are from others: memetics and mimesis. Almost always the conventional wisdom will be better than what I can cook up myself, and it’s less costly, so why should I do anything other than copy? Even people on the frontier are only on one, or Maybe a few frontiers. It’s extremely rare for someone to be on multiple frontiers: it’s hard enough to be on one!
In closing I’ll say I was surprised that the final two twists of the book were, first, the AI was running the vampire the whole time, and then second, that the vampires took over back home. It seems more plausible to me that if the AI is smarter, and it’s all about smartness sans consciousness, the AIs should’ve taken over
Looks really good. Need more interesting and useful AI UIs
We built Claude Code into a browser & whiteboard! It's called Decode @decodetool
It’s the fastest & best way to give UX feedback to Claude Code while coding locally & enables it to review & test changes itself
See thread👇to learn how whiteboards makes it an even better browser
