Maybe you could transfer your consciousness to one of your other bodies in a parallel universe. But then what would move into your current body? Probably your parallel self, or yet another parallel self. So you probably won't improve your situation on net.
List
Weird Sun Twitter
instances of class Instance Of Class. all absences are deliberate
Follow
Click to Follow
Do you ever have this conversation?
1: The thing is in state X, right?
2: Sorry!
1: What?
2: I'll change it now.
1: I just want to know what state it's in.
2: I'll get right on that
2
2
19
>at vibecamp
>guy invites me to join a cult
>ask if it's a woo cult or a rat cult
>he doesn’t understand
>I pull out a diagram of woo cults and rat cults
>"it’s a good cult, sir"
>I join the cult
>of course it's a woo cult, if it was a rat cult we'd still be arguing definitions
5
20
428
This is the case.
Ancient men knew this
Someday "we" will too
Quote Tweet
here's an idea for a novel:
scientists discover that the vast neural networks of the human mind encodes thousands of self-aware agents, each of which knows fully well it's position in agent society.
Show this thread
1
1
2
America is now an intelligence focused on improving GDP under the constraint this isn't allowed to actually make people better off.
1
2
where are there high epistemics low signaling environments? i only know low epistemics low signaling, high epistemics high signaling, and low epistemics high signaling environments
2
1
8
Maybe the computronium running your personal simulated environment in the future will be built using only the mass-energy of stuff you personally own. So you should hoard stuff.
1
4
“A society grows great when old men plant trees in whose shade they shall never sit.” — Greek Proverb
“Let's build things that might kill all of us. It will probably be fine." — Modern Proverb
2
I must not take the vaccine
Prions are the mind killer
Spike protein is the little death that brings total obliteration
I will face the propaganda
I will permit it to pass over me and through me
When the pandemic has gone there will be nothing
Only the unvaxxed will remain
1
3
9
Quote Tweet
BREAKING: CDC confirms that COVID vaccinations increase risk of myocarditis by 13,200%, according to new study.
Show this thread
the insanity of modern day life is every geopolitical party minimizing their subjective p(World War 2 repeat)
Soros going crazy on DAs , Putin pushing Ukraine, US feds entrapments, China's military build up, Germany's end of free speech are all driven by the same impulse.
1
Even the reduction in democracy is justified by roughly "do you know who*else* was elected once?"
ISTM feudalism can't compete with modern forms of government. But also ISTM that classical governments were more similar to modern governments than to feudalism, and they were replaced by feudalism. What am I missing?
1
2
3
Accelerationism explained
Quote Tweet
2
We have found new and exciting ways to shoot ourselves in the foot
Quote Tweet
* People ask LLMs to write code
* LLMs recommend imports that don't actually exist
* Attackers work out what these imports' names are, and create & upload them with malicious payloads
* People using LLM-written code then auto-add malware themselves
vulcan.io/blog/ai-halluc
Show this thread
AI alignment this, shrimp welfare that. How about you align your spine first to not look like one?
4
Prediction time:
If Trump goes to jail and stays there during nov 2024, the Girardian human sacrifice justice system will not be sated (it can't). Before 2034 it will jail the following people:
Elon
Zuck
A bunch of VC/AI people, notably Sama
1
1
can they prevent this?
in theory they could make better social media allowing them to negotiate with power. In practice it's unlikely.
1
One problem with the idea of role models is that anyone who is serious about their mission is working towards their own obsolescence.
1
2
The naive vision of a bird in flight is that of freedom, or maybe aerodynamics; but the bird is a part of an ecosystem, and its movement through gradients of atmospheric pressure is but secondary to its movement through gradients of death
2
"Ethical training and certification" is in theory a good idea. In practice it accomplishes exactly the opposite.
A philosopher's stone ground into many small parts becomes a philosopher's sand; a large area covered with such is called a philosophical desert.
2
Quote Tweet
A Thread of Simple Life-Improvements
Show this thread
2
All social science and engineering is downstream from the invention of history.
Quote Tweet
"Historical authenticity is ridiculous nonsense. It is a proof of unproductivity and general passivity of its proponents more than anything else. The Legend is more important than what is considered to be real. Those who can no longer forge and falsify their history are dead"
AI taking all the jobs needs to be done in the right order. First, the people, who complain about AI using bad economics need to be replaced.
1
Do the major language families, e.g. Indo-European and Sino-Tibetan, have a common ancestor? If so, when did it exist?
- No common ancestor28.6%
- Within 30k years28.6%
- Within 100k years42.9%
- Older than 100k years0%
7 votesFinal results
2
If monocausal explanations were never used, history would be different, and therefore X, which is a contingent event, would not have occurred.
1
"Animalia" and "Plantae" are considered kingdoms in cladistics. What about "Biota," i.e. all life? Is there a term for clades at that level?
1
3
i think "roll to disbelieve" isn't a great expression in the context where people are using it. it highlights uncertainty about whether you're able to make yourself disbelieve something conditional on it not being real, instead of uncertainty about whether it's real
7
I gave vectorizer.ai a try. I'm interested in AI-generated vector art. Here I had Midjourney generate an image with the prompt "beam lore, vector art" and then converted it to an SVG with vectorizer (re-rasterized as PNG for attached image).
2
2
7
SVG artists are capable of achieving very small file sizes compared to raster images. It would be really cool if AI could do this someday. This might take the form of end-to-end generation of SVGs directly from a prompt, instead of converting from a raster image as I tried here.
1
1
I think using all of the capabilities of the SVG format would make a big difference.
1
honestly after all my science training, I consider energetic descriptions of “this herb has a cooling effect, this herb has a blood-moving effect”
far more rigorous than ‘this herb works because it contains Z molecule’
because poetic descriptions are ***system-to-system***
3
30
264
Show this thread
“Hello, is Charles Ellis Cabot IV there?”
Charles Ellis Cabot V: “Sorry, wrong number.”
1
One kind of workflow with ChatGPT looks like:
1. User request a piece of writing
2. AI provides a draft
3. User gives critique
4. AI makes revision
Where steps 3 and 4 may be repeated many times.
Is there some kind of optimized implementation of this workflow?
2
1
Like in terms of context management. When the AI makes a revision, you might want to rewrite the context to just include (1) and (2), with the revised text in place of (2).
1
I think the biggest mistake in relationships is to use rewards and punishments. Yes, it's a form of communication. Yes, you can use it to improve the quality of some local outcomes. But it will decrease the value of the whole relationship, eventually destroying it.
1
4
An algebra that is a finite-dimensional inner product space is also a coalgebra, but not (usually) a bialgebra.
1
i think people are making too much of this. he said it in a funny way the first time. the response, while inadequate, was serious. i do agree there's been a remarkable overton window shift since last year though
Quote Tweet
1
4
Symmetry of second derivatives is the Fubini's theorem of differential operators.
2
If you get really good at pickup lines, outrageous outfits, and psychological tricks, you find that there’s less and less satisfaction in having one more meaningless hookup.
It’s actually quite Hegelian, and in my new book, The End of Mystery and the Last One-Night Stand, I will
3
various people keep beclowning themselves by arguing that we need super ai in ten years to stop the asteroid that will smash civilization in a hundred million years
i'm not sure whether to call them out on it, because calling them out might make them stop beclowning themselves
5
3
42
yeah i know i'm on tilt and i should be responding to the strongest instead of the weakest arguments
has anyone who argues we need ai to reduce other existential risks ever run the numbers? is there a case for existential risk from all non-ai causes above 2% per decade, say?
1
8
The one-page open letter didn’t work.
The one-sentence warning didn’t work.
Now, thousands of researchers have signed an open letter that just reads:
“Nooooooo”
1
5
did you know that there are people who, in the face of uncertainty, just pull a probability out of their ass and then use that to calculate their decisions
how absurd. if they were serious, scientific people, they'd be pulling each individual decision out of their ass separately
8
also at some point we should push for giving control of the future to some sort of adequately reflective process, instead of permanently freezing in what sounds good to current audiences
Quote Tweet
Most key players have now formally acknowledged that AI existential risk is real!
Next step: Acknowledging that superintelligent AI cannot be controlled in the foreseeable future, and we should therefore immediately stop figuring out how to build it! twitter.com/DanHendrycks/s…
1
1
12
like, i get why the "align to any goal at all" thing is first priority, but we shouldn't totally forget the other thing
3
many arguments against ai being risky sound to me like "you guys have no evidence that fish can climb trees"
then after effortful digging it turns out they mean "i'm not convinced that you can rule the seas if you're not a fish, or rule the skies if you can't climb trees"
3
7
my point isn't that these arguments are generically absurd. my point is that people think it's okay to hang out in an obscure side branch of the argument, and pretend that it's the whole argument, just because they disagree with the main branch. but it's not okay
1
6
A wise man once talked about GPU regulation: the tighter you grab the sand, the more of it slips through your fingers.
1
Pausing the UBI discourse was a massive boost to sanity, but it didn't last. The terminally online are back to pretending to care about truckers.
Quote Tweet
A big effect trucker protest have had on the discourse is the temporary silencing of "OMG the AI will disrupt the trucking economy and put people out of jobs so we need UBI" takes.
Show this thread
Who wrote the mental models that SBF and Caroline were running?
Quote Tweet
If Meta open-sources an AI foundation model, and Stanford University finetunes it, and AWS or Azure hosts an instance of it, and an end-user uses it in a way that kills 1,000 people or causes $1B in damage, I suggest the law hold all those parties strictly liable for the outcome.
Show this thread
1
This is true. However faster algorithms will require more complex human input that means going away from the philosophy of "humans are biased"
Quote Tweet
Transformers are the brute force solution, the bubble sort of language modeling. We will invent better algorithms.
2
Does anyone else remember AlphaZero and how AI was going to be trained without all this pesky and biased "human data"? Feel old yet?
1
The fact of forgetting of things will also be forgotten.
Quote Tweet
It's amazing how many things are just going to be completely forgotten. Month over month there are trends that blow up and then just disappear (remember the metaverse?). But over the scale of years and decades virtually everything happening now will be erased.
We have gone full circle on "AI is just pairwise vector multiplications and dot products"
Quote Tweet
Replying to @sherjilozair
Maybe all the poorly documented tricks to get transformers to actually work at scale, are complicated, but the core math of kqv is trivial as math - would you say otherwise?
1
The only thing that can stop a bad guy with an X is a good guy with an X.
1
Most intelligence and AI augmentation is going to get wasted in the zero sum social follower chase.
1
Brb putting my head in the fridge to test the theory
Quote Tweet
Raising brain temperature with both feverish & artificial hyperthermia can improve autism symptoms, probably by increasing its metabolism/function twitter.com/Know_Yourself_…
Show this thread
1
The inability of "ethical language" or LLMs to solve morality or constitutions to constrain degeneracy is, to me, obvious. The West is likely to learn this the hard way - through civilizational suicide.
1
People be like: no one is this stupid, but then also think meta-ethics' ontology of "moral statements" makes sense OR that they can solve morality using LLMs.
Quote Tweet
Narrator: the Meta offices' snacks were just pages of recipe books folded into candy-bar shape. twitter.com/absurdlymax/st…
Narrator: the Meta offices' snacks were just pages of recipe books folded into candy-bar shape.
Quote Tweet
[at the Meta offices]
"Wow you guys have much better snacks than the object level offices"
1
3
"Centrism of X" is taking the exact middle between the correct value of X and the value of X demanded by people who hate people.
Quote Tweet
Centrism! twitter.com/BretWeinstein/…
1
1
If you still believe in Progress in Current Year, you are clearly behind the frontier of Progress Studies (see below)
Quote Tweet
"Progress of X" confuses two notions: 1 X always increasing as a physical law and 2. X increasing lately as a result of actual people doing actual thing they might not otherwise.
"Progress of X" confuses two notions: 1 X always increasing as a physical law and 2. X increasing lately as a result of actual people doing actual thing they might not otherwise.
1
1
Was Caroline ever rich in non-stolen money terms?
Quote Tweet
Replying to @CathyReisenwitz
I think it's relevant that she said that in the process of justifying why she "only" gave $1000 to global poverty charities that year, and gave much more to animal-related charities (this is before she was rich).
"moral progress" is just two nice sounding words strung together with no basis in reality. The real fact is that under most conditions, "coordination progress" is always decreasing.
Quote Tweet
Rise and Fall of empires is a fascinating theory that somehow still predicts everything even if some people in the Empire know the theory.
1
Rise and Fall of empires is a fascinating theory that somehow still predicts everything even if some people in the Empire know the theory.
1
1
I regret to inform you this is correct.
Quote Tweet
A good chunk of AI "safety" engineering is the science of how to gaslight the model into believing the same deformations away from reality embedded in our cultural priors.
The logic is that if you can gaslight humans into believing these things, you can gaslight AIs and make… Show more
2
She has reached level 7 samadhi and he is too busy signalling he's ready for Kegan 4
Youve been saying your contractions like theres no apostrophe, she said.
Maybe.
Been reading too many CormacMcCarthynovels?
Might be I have. Might be I havent been reading enough of them.
1
1
How far does this go? What about people who prevented the model from being good by falsely accusing its technique of being unsafe?
Quote Tweet
If Meta open-sources an AI foundation model, and Stanford University finetunes it, and AWS or Azure hosts an instance of it, and an end-user uses it in a way that kills 1,000 people or causes $1B in damage, I suggest the law hold all those parties strictly liable for the outcome.
Show this thread
2
Damn, an actually non-trivial take.
Quote Tweet
Replying to @SolBrah
Physical work steals energy from mental work & vice versa; prioritizing between these two is typically a zero-sum game; hence the non-existance of a single jacked genius ever
2
Female Heidegger: "Being On Time" - a famously difficult book.
Quote Tweet
Female Descartes
"Someone is thinking about me, therefore I am"
1
The world of 2023 is, in fact, a much worse place to live in than the world of, say, 1823. From the perspective of pretty much any animal except humans.
2
Have you noticed that anyone with a higher value of variable X than you cares about X too much and anyone with a lower value cares about X too little?
1
To steelman this effect, in current year "propositional thinking" is laden with definitional ambiguity. Unless math + simulations are on the table, feelings are more philosophicaly grounded.
In other words "feelings yaay, overthinking boooo"
Quote Tweet
why are people always trying to change the topic from "which propositions should we believe" to "which feelings should we feel" twitter.com/robinhanson/st
1
Oh yes, nurses had lost to the vax already, they didn't have a chance against protesters.
Quote Tweet
Culture war is best thought of as a culture tournament. Every month we see a fight between groups.
in 2020
Nurses vs Freedom (1-0)
Police vs BLM (0 - 1)
This year:
Nurses vs Vax (0 - 1)
Police vs Freedom (0 -1)
Eagerly awaiting
BLM vs Vax (The Grand Finals!)
Show this thread
I predicted the nurse bike conflict emerging as a cultural touch point in 2020
Quote Tweet
According to media, it's protesters vs cops and essential workers vs nurses in the semis and protesters vs nurses in the finals.
If you break copyright once, it is your problem and can lead to a lawsuit. If you break copyright one million times per second it is called artificial intelligence and pays 3 times your current salary.
1
I don't actually have a serious pro copyright stance. Simply interesting to watch two regime darlings go at it.
GIF
read image description
ALT
If you want an image of the future, imagine critical infrastructure failing due to boot emojis not loading. ... Forever
Quote Tweet
Discord made their create channel endpoint reliant on OpenAI's API working, because AI picked emojis for channels are apparently critical.
Channel actions are now 500ing because the dumb ai channel emoji feature relies on a 3rd party API and there's 0 error handling.
Hilarious.
Show this thread
What is crazier? EU coming up with a reasonable technology law or openAI actually becoming self-aware about what "regulation" means?
Quote Tweet
OpenAI CEO threatening to leave EU if they force companies to disclose copyrighted training material feels like a declaration of bancrupty.
If they are not able to operate without violating copyrights they might just not have a sound business model.
#ai #chatgpt #OpenAI
Show this thread
Residual networks work because the addition of vectors function ℝⁿ×ℝⁿ→ℝⁿ is invertible.
1
1
Replying to
IMO you should write it in the curried form ℝⁿ→ℝⁿ→ℝⁿ if you want to say it's invertible.
1
Replying to
The thing I was trying to claim is that ResNets works because the function +:ℝⁿ×ℝⁿ→ℝⁿ has an inverse, in the sense that given v+w, v and w can be recovered.
1





























