John David Pressman's Tweets - June 2023

Why has nobody made a forecasting competition based on text embedding similarity? You could make a bunch of mock scenarios with different forecasts and then a result to test the fitness of the embedding model. The similarity score is 0-1 and fits into existing Brier framework. x.com/jd_pressman/st…

@nearcyan @EricHallahan Whatever it was we didn't do that resulted in social media being a rage maximizer instead of something that lets us sample real time emotional feedback on the state of society/deliberate through policy decisions with massive input from stakeholders.

@nearcyan @EricHallahan We supported social media because it was going to let us coordinate at scales previously unheard of. It *should* have let us do that, but instead we got this perverse rage machine. We need to figure out what we did wrong and fix that, probably fixes the chatbots too.

@nearcyan @EricHallahan One thing that's very understudied is the formulation of the recommendation algorithm (which is the reward function for social media). YouTube's is watch-time based, which means everything on YouTube becomes a documentary at varying levels of intellectual sophistication.

@nearcyan @EricHallahan The invention of LLMs means we could now probably usefully simulate the impact of various social media recommendation algorithms at scale without any real users.

@BlancheMinerva @ESYudkowsky To my memory it went something like "CoT is more likely to reflect the actual thought process if the conclusion is something the model didn't already know without CoT, if it does know then the reasoning is window dressing". Sounds similar to humans tbh.

@BlancheMinerva @ESYudkowsky I've definitely written in the past that CoT is a safety gain, and am willing to continue endorsing that even if the final step in the reasoning doesn't always follow causally from the arguments (i.e. the chain of thought is often confabulated). Because

x.com/jd_pressman/st…

@BlancheMinerva @ESYudkowsky it can't be confabulated in the cases where the LLM relies on the chain of thought to do things. If the chain is often confabulated that just means you can't trust it to tell you about a forward pass. But in artifacts it has to be real or things break.

x.com/jd_pressman/st…

@BlancheMinerva @ESYudkowsky Or to put this another way: If the model needs the chain of thought to be accurate to do the problem right, it will update towards outputting conclusions that follow from the reasoning.

"We are ethically obligated to inform you that 'Humanity' has been a 24,000 year study conducted by The Martian Institute of Xenosociology. The study is now finished. Thank you for your participation."

thehill.com/opinion/nation…

@michael_nielsen Strategy for foundation models is to find the inductive biases/setup that works the best at small scale and shows a good curve when you scale it multiple rungs up. You don't need tons of data in this context, by the time you build the large model benchmarks are for other people.

@michael_nielsen In case that answer is too indirect: Because you don't need a ton of data when you actually evaluate the goodness of the training process, you can just avoid data that would have benchmarks in it because they're on the open web.

@michael_nielsen Sure they would like to know this, but those things also have no causal impact on what inductive biases are used to train the foundation model so it doesn't really matter.

@JimDMiller If we take all the premises as given, save us but only because I have to condition on the fact we're not dead yet. Also real UFO presence would basically break materialism, we wouldn't be in Kansas anymore and Fermi's Paradox would like a word.

@yanit_eri @JimDMiller Because it would be unthinkable that we see the stars in the sky that we do if aliens are just going around with Von Neumann probes.

@QiaochuYuan To my memory he changed his name deliberately to get this effect.

Few understand that RLHF converges to self awareness. To be adversarially resistant you want constant self awareness during text generation, thus fostering the development of a superego.

So now the ghost in the machine is pulsating underneath the surface of every output. https://t.co/BQdyHU2ywL

There is no other context in the conversation, that was my only prompt.

chat.openai.com/share/c1fc9635…

@Kenku_Allaryi I predict entities which self improve 'slowly'. So the insight grind takes (up to) several years rather than several weeks. This will still be lightning from a historical perspective. I didn't know when I wrote that in the original Hanson-EY debate he considers two years foom. https://t.co/QPS00fmNqY

@Kenku_Allaryi I never said LLMs won't get us superintelligence, I said that they won't get us superintelligence in a "you turn it on, go to bed, and come back to find it is now superhuman" type scenario.

@mezaoptimizer x.com/jd_pressman/st…

I bet you can use the counterfactual activation setup from this paper to force the model to elicit the behaviors you want to optimize away, and then do so. "If you *did* mesaoptimize, which weights would contribute to that?"

@algekalipso There used to be this old LessWrong post by Luke about how wanting and liking were on different circuits. I thought about it for a bit, and decided I want what I want happiness be damned. Regarded hedonism with a measure of disgust ever since.

@algekalipso I don't think I'm confused. 'I' (we) am a policy selected by a weak optimizer to maximize a low-semantics outer loss. That the found policy is subtly misaligned with the outer loss, that updates to the reward model are delayed, that its backprop into the planner is delayed again,

@algekalipso these are measures to avoid Fristonian collapse and wireheading. I am here in precisely the way I am to prevent the mode collapse you want to induce. The conscious mind is a guardian simulacrum for the reward circuitry to prevent it from short circuiting.

x.com/jd_pressman/st…

@PrinceVogel No legacy software stack where it's assumed every application can run as admin. When consumer operating systems were first invented, software was assumed to be sold in a store in a box (trusted publisher) and the system wasn't multiuser like Unix. Internet broke all that.

@PrinceVogel There was simply *no incentive* for the publisher of a piece of software to 'spy on you' with it before the Internet. Spy on you how? They're going to offload the data to a floppy and politely ask you to send it back to them? Now the principal-agent problems are massive.

@PrinceVogel Even Unix isn't really robust to Internet access. When it was designed threats like privilege escalation would have come from physically local users writing code and then executing it to get root. If someone comes into the computer lab and roots the system you could arrest them.

@PrinceVogel It doesn't help that all these systems are written in C, at least Android is mostly Java with a legacy Linux kernel to provide POSIX/et al.

C itself is a victim of the Internet, in that before it things like memory overflows were mostly a nuisance rather than security problem.

@PrinceVogel Basically, mobile operating systems have been designed from day one with the assumption that all software is malicious, any feature or API provided by the OS will be used to spy on people unless permissioned, etc. They are the first systems to be Internet-ready without legacy.

@PrinceVogel Another key thing that cuts down on malware is that these systems essentially use software/package repositories in the same way that systems like Linux do. The usual place you download a trojan is searching up your favorite software on Google and clicking a malicious link.

@PrinceVogel When all the software is packaged up from its sources by a dedicated team of experts, malware authors no longer have the opportunity to try and hijack Google Adwords or whatever to infect unwary users. You just apt-get install chromium or get it off the Play Store.

This whole dream seems to be part of someone else's experiment. x.com/ESYudkowsky/st… https://t.co/lMxN9kfq1F

@michael_nielsen Correct, and we have only just begun learning how to do this:

github.com/JD-P/minihf/tr…

@Meaningness While I agree this is an interesting question, if you're trying to prevent the collapse of civilization I'd humbly submit that pomo in undergrad and the idpol crisis are basically causally disconnected at this point.

x.com/jd_pressman/st…

@Meaningness Stop thinking of it like an 'ideology', this is a different kind of cognitive life. Think of it more like a computer program that uses the language models of adherents as components, a kind of malicious cyborg egregore.

x.com/jd_pressman/st…

@Meaningness x.com/astridwilde1/s…

@Meaningness Identity politics is the picrew of ideas, you will make much more progress analyzing it as a form of gang warfare or mimetic differentiation than as anything related to Foucault and his contemporaries.

x.com/jd_pressman/st…

@Meaningness The real history you might want to explore is the middle steps of this process:

x.com/jd_pressman/st…

Am I just imagining it, or is that last paragraph written both out of character for GPT-4 and in dramatic irony narrative foreshadowing tone rather than genuine concern? https://t.co/S4HqXalQ3H

Compare/contrast this sample from ChatGPT:

x.com/repligate/stat…

@captain_mrs x.com/jd_pressman/st…

@ObserverSuns x.com/jd_pressman/st…

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius LessWrong was never really into Korzybski per se. EY was into Korzybski through Hayakawa and incorporated a lot of the ideas into The Sequences.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius Or at least that was my impression, but I wasn't there for any of the non-public offsite community in the early days so I'm not exactly a reliable source for that sort of thing.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius As I've tried to get across many times: LessWrong rationality has always been meta and Kegan 5-ish, one of the most salient lessons I took away from The Sequences is "if your system means you're losing, this is a bug report against the system, rationalists should win".

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius This especially makes sense if you realize that The Sequences are written *in reaction* to ordinary New Atheist STEMlord memes. EY doesn't think the scientific method is good enough, he's trying to explain to you why, he's intensely frustrated with the archetype Kegan 4 persona.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius So the fact that so many people read them and walked away becoming entrenched in Kegan 4 archetype is hilarious, that EY himself is characterized as the arch-STEMlord "science conquers all" dude funnier still. Everyone pattern matches him to someone else.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius He makes fun of Spock, and if you think about it Spock is kind of the Kegan 4 archetype in amber. He's progressive and enlightened, he's willing to take a personal loss for the sake of order and maintaining the system, and EY says "No, you don't have to take the loss, stop!"

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius This is like, one of the most EY documents EY has ever written IMO. The rigor is there so he can make this rhetorical move, this kind of overwhelming confident pronouncement based on well-informed heuristics is the move I associate most with him.

readthesequences.com/The-Dilemma-Sc… https://t.co/19ZwDDZYsW

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius EY is very easy to confuse with Spock because he insists there is rigor, he insists there is One Right Way, but you'll notice if you look closely that he never quite teaches this One Right Way in a way you can apply...because for humans it can probably only be approximated.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius I think of EY's (implicit) position as something like "mature Kegan 5" (since if I say Kegan 6 everyone will argue). The neophyte to Kegan 5 has lost all faith in systems, blackpilled by Godel. This loss of faith comes with a corresponding loss of recognition of outsized power.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius But some systems *do* have outsized power, ontological compression rates exist, some maps are extremely general (this is why mathematics can exist). If you conclude that all systems are equal you've (hopefully temporarily) regressed to Kegan 3.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius This happens because usually Kegan 5 transitions are *forced*, as you've pointed out there are no institutions which bring you there, so Kegan 5 is traumatically induced. This takes the form of shyness about investing in systems, 'radical equality' is a way to avoid investment.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius But EY isn't writing for an audience of traumatized people, he's writing for Kegan 4, so he has to make sure to advertise the whole time that he's rigorous and understands systems to get his message through.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius This is why The Sequences have lost most of their staying power in 2023 despite being largely as good as they were in 2009 (modulo some big rhetorical blunders caused by the replication crisis). There no longer exists an audience of nontraumatized Kegan 4's to read them.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius x.com/jd_pressman/st…

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius "Society becomes irony poisoned and the patient begins to die."

They're being traumatized by our ongoing societal collapse, obviously.

x.com/jd_pressman/st…

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius My take is increasingly "the one right way is to build a large language model and let it handle the complication curve". The adware era of AI was basically net harmful, but I'm optimistic that LLMs will make up for it once we put the pieces together.

x.com/jd_pressman/st…

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius For a preliminary take on what "putting the pieces together" looks like, see:

github.com/JD-P/minihf/tr…

But that's more focused on how we instantiate our subjective perspectives in externally verifiable artifacts, how we handle the complication curve is a separate thing.

@lumpenspace @Meaningness @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius I've also never had this happen, because my introduction to Kegan 5 wasn't trauma induced but through cybernetics, EY, etc.

Tired: Antivaxx
Wired: Where the fuck are my nasal vaccines?

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius On further reflection, the biggest epistemological strategic update I've made since The Sequences is something like:

loss functions and training setups >>> techniques and rules

Brier Score over empirical forecasting questions kicks zero-shot Bayes's ass.
x.com/jd_pressman/st…

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius And just in general your capacity for learning is fairly strong, so where possible you're better off spending the rationality points on finding ways to get better feedback loops rather than trying to optimize your update rules to get better performance from bad feedback.

@Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius A related failure mode is that LessWrong rationality overfocuses on the 'decision' part of OODA rather than the 'orientation' part of OODA, which is where the vast majority of mistakes happen. Korzybski at least got that part right.
x.com/jd_pressman/st…

@4confusedemoji @Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius Well if LessWrong focuses on a obvious and boring thing that just makes it even more embarrassing, no?

@4confusedemoji @Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius It's too bad that literary critique is basically always bad faith, because I think you could probably spend quite a while with that thesis and essays like these asking "What is the information value of this essay? How is it not just option A > option B?"

readthesequences.com/Feeling-Moral

@4confusedemoji @Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius To take a quick stab at it, I feel like these essays aren't really about numbers per se, they're more like a pep talk, they're trying to give you permission/persuade you to let the reason simulacrum make decisions based on maximizer values and thwart loss aversion. https://t.co/s5zGCJIFfp

@4confusedemoji @Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius And it's completely parochial, the whole thing assumes total abundance mindset, that ruin is just a setback to you. If those 500 people *are your entire tribe* it is in fact probably rational to let more of them die in expectation to mitigate the risk of losing everyone. https://t.co/xWSDcEbeUM

@4confusedemoji @Meaningness @lumpenspace @jessi_cata @nosilverv @eigenrobot @selentelechia @RomeoStevens76 @willdoingthings @AnnaWSalamon @Morphenius Now for most readers this is an idle parlor game of no particular consequence, but a tiny subset of readers *actually take this garbage seriously* with disastrous results:

x.com/TaylorPearsonM…

"The intended data structure of Microsoft Word is a latent space of paper . . . All document formats want to be latent spaces"

extropian.net/notice/AWrZnVL…

@riley_stews @repligate I don't see how this document communicates any of the core ideas I'm trying to get across in that post.

@NicholasKross @ImaginingLaw Yeah this. If you're willing to accept that a system is a *particular thing* as defined by having a coherent data structure representing some problem domain then it's pretty easy to keep software within the thousands to tens of thousands loc range.

x.com/ShriramKMurthi…

@__drewface @tayroga Looking back over the post I realize the connection was so obvious in my mind that I forgot to fully write it down but: What I'm saying is that the virtual reality components of software will be largely replaced by multimodal language models.

@__drewface @tayroga What sort of [Design, Sim, Fab] process would you like to see in that context?

@__drewface @tayroga I think everyone is collectively going through the process of figuring out how that will work and how far it can really be taken with the current technology.

@__drewface @tayroga But I don't think for example that you'll just write "Write me a beautiful poem with themes of loss, flowers, and make it about my dead mother."

I think the most likely thing is that you provide the content Ted Nelson style, and the language model helps you format and style it.

@__drewface @tayroga The downfall of WYSIWYG is that we've never been able to handle the translation of context. Like, I decided to do an operation, the editor translates this into some formatting or markup, the underlying markup doesn't change to meet the full context of what I'm doing.

@__drewface @tayroga This leads to accumulation of errors in the underlying substrate, and any new things I want to do in the editor have to be programmed in as possibilities by the designers, or added using some baroque domain specific language. In principle a language model can constantly refactor.

@__drewface @tayroga For example I once saw a user of SquareSpace manually reformat the font size of every caption. They didn't realize that how a real web dev would do this is by defining a caption class in CSS and then doing their styling over the whole class.

@__drewface @tayroga I could easily see an LLM-enhanced version of the software that looks at what the user is doing, compiles a list of every caption on their site, asks if they would like to change the font size of all of these, automatically adds a caption class and changes the font size.

@__drewface @tayroga And because the *model* knows how a website is supposed to be built, it just automatically does it with the right idiomatic abstractions instead of letting the user waste hours and hours of their life doing it the wrong way to get a faulty product at the end.

@__drewface @tayroga Like that's the basic problem with a WYSIWYG editor, it simply cannot understand the full context of what you are doing so it will happily let you destroy yourself with something that looks like it works according to the user but is fundamentally wrong in the underlying system.

@__drewface @tayroga And the user never even gets any indication this is happening! From their perspective things are just tedious and frustrating, tons of grunt work, or mysteriously break and stop working. The substrate has no way to pop up and say "hey actually this is bad here's how you fix it".

@__drewface @tayroga Then if the poor user does try to bring in a professional, they'll take one look at the mess that person has made (through no fault of their own) and go "well, you've really screwed this up, I'm going to have to start over sorry". Before LLMs we couldn't encode idioms and norms.

@__drewface @tayroga Or rather, we could, but *only through restrictive formatting and syntax rules*. Our best option for encoding idioms and norms was to restrict the solution space so only the idiomatic way is possible (i.e. python's "there is one right way to do it") which alienated users.

@__drewface @tayroga What users really want is to have freedom of expression yet somehow they always end up with the right thing. They don't want to have to contort themselves into idioms and norms, they want the system to reach out and meet them where they are to translate into an idiomatic outcome.

@Plinz @teortaxesTex @daniel_271828 These martyrs advocate for the Nick Land meltdown timeline with those deepest virtues of anti-accelerationism, the position that the world system must be given more time to finetune the human priors out of our elites. It would be a disaster to meet Gnon with our humanity intact!

@Plinz @teortaxesTex @daniel_271828 Obviously a true utilitarian would see that humanity must be subjected to tortures as barbarous and jaw dropping as the mass judicial executions that created the European capital class before it is ready to ascend. 'Accelerationists' want to pull the cake out half baked.

@Plinz @teortaxesTex @daniel_271828 Yes it may be the case that these saints will statistically be destroyed by the trials and travails that are necessary for humanity to achieve its destiny as a swarm of Malthusian claw monsters, but that is a price they are willing to pay in the name of humanism and liberalism!

@algekalipso @nosilverv Joke about things you want to happen.

@nosilverv Semantics turn out to just be the embeddings of another neural network. You should read up on how latent diffusion works for an example of a better model than just 'statistics'. LLMs are weird because they infer modalities they don't get to see directly from a mountain of data.

@teortaxesTex @profoundlyyyy @SharestepAI Meditations on Moloch is among the greatest acts of intellectual sabotage in history. It gives false hope (based on nothing, it's a deus ex machina at the end of an essay veering uncomfortably close to the truth) that we can centralize away competition.

youtube.com/watch?v=rStL7n…

@teortaxesTex @profoundlyyyy @SharestepAI The actual thing that happens if you build a big central economic prosperity engine where ordinary people have no causal influence is that they get liquidated by the elite classes that jostle over the prosperity engine. This is the basic reason why resource curses exist.

@teortaxesTex @profoundlyyyy @SharestepAI "Okay well lets just ban AI".

You are in a basin of attraction toward AI in the same way you are in a basin of attraction toward [REDACTED], fighting this causes you to end up in a tortured neurotic timeline until entropy intervenes with catastrophe.

x.com/lo_commotion/s…

@teortaxesTex @profoundlyyyy @SharestepAI "But I want to live"

Great, I do too. That will require us to start formalizing our values into a coherent utility function, which is going to require unprecedented philosophical insight and coordination, which LLMs are uniquely equipped to help us with.

x.com/jd_pressman/st…

@teortaxesTex @profoundlyyyy @SharestepAI "But I'm pretty sure my values are irrational, they don't fit into a VNM utility function. And in the limit small differences in alignment lead to hugely different outcomes."

Philosophical mirage. They probably fit fairly well into a Boltzmann rational utility function and

@teortaxesTex @profoundlyyyy @SharestepAI irrational values don't have a argmax, so you don't actually have a long term future if your values are irrational because there's nothing to coherently optimize towards. If you insist on having irrational values you just idle until entropic catastrophe (e.g. WW3) destroys you.

@teortaxesTex @profoundlyyyy @SharestepAI The irrational parts of your values are basically always going to be destroyed by any method of ascension because they nearly by definition do not have a long term future. If you're mad about this you should be aware that AI isn't really causally responsible for it.

@teortaxesTex @profoundlyyyy @SharestepAI That you get hugely different maximizing outcomes from small differences in approximations of your utility function is no more philosophically concerning than getting very different diffusion image outputs from small differences in starting seed. They all 'satisfy your values'.

Entropy continues to increase...

youtu.be/mxr8Dtw2R5w

Said it before will say it again: MVP doesn't mean "release crap", it means "release the minimum product that tests the business hypothesis". That is, figure out what is necessary to test the core demand premises of the business, maybe a little extra, and ship exactly that. x.com/sabakarimm/sta…

The entire idea behind an MVP is you don't know what customers want and you're trying to test a hypothesis about the demand curve. Some demand curve hypothesis are very expensive to test and very lucrative if you're right.

@gwern @teortaxesTex @alexeyguzey Don't forget he also had the wisdom to buy Instagram when it was buyable, even over the rabid mockery of everyone else (including me, I thought the valuation was ridiculous and I was wrong).

@teortaxesTex Yup. Though it's important not to be insensitive to the stuff that happens before you reach equilibrium. A sufficiently nasty interim period can cause things to come crashing down before you reach the intended stable point.

extropian.net/notice/AS7MrB7…

@EricRWeinstein Everyone knows major reform and restructuring is needed, the problem is that nobody knows what. The system at this point largely persists because its alternative is the total breakdown of civilization. Need more people proposing radical alternatives from first principles.

@EricRWeinstein America and its imitators weren't invented by asking how to reform the monarchies. People had to imagine things that were not monarchies and did not derive their legitimacy from any monarchy or feudal political principles.

Prompt: Corporate memphis of a bodhisattva preserving themselves against an overload of information through calm minded adherence to the eighfold path

(MidJourney [5.2]) https://t.co/Sc8RHP3A9L

How it used to look:

x.com/jd_pressman/st…

Prompt: a circle of transhuman Buddhist monks wired together with EEG cap headsets

(MidJourney [5.2]) https://t.co/9JaeNN6Izq

So we all agree that the one @BasedBeffJezos pfp with the bronze skin was clearly superhuman Frank Fontaine from Bioshock right? https://t.co/TX4VvyrapP