John David Pressman @jd_pressman Twitter profile

Pinned Tweet

John David Pressman

5 months

MiniLoom nearly ready. Settled on capturing user data from three sources: - Which nodes a user chooses to expand (proof of value through child nodes) - Diffs between AI generated nodes and user rewrites (what specifically was wrong) - AI rewrites from bullet point feedback

4

1

45

Last Seen Profiles

@PoliciaSXXI

@paulkats265

@WelbourneStud

@REALDJPREMIER

@illuminatibot69

@lyricthompson

@Slap_BattlesRBX

@Junichi6969

@Aza97181937

@ValkAdx

@gyrofield

@taetongas

@maiasalvo2025

@TheSafeAnnaAnon

@YehyaAmmer

@AtlanticFBO

@FerraioloJ65413

@outi_veivo

@ZAPAL118

@iamzahryamaee

@TMartinDeng

@Lyndenboy

@Neauso155865

@Faizytofficial

@BenBelly4

@Lubi_Mac

@p7sports

@BL3P

@TCKievBE

@AkterHusne92605

@johnsummit

@Kitakartel

@rugiyyamohamed

@Cobrabil2023

@CrowdCompass

@BronzeWhale

John David Pressman

@jd_pressman

2 years

What I thought OpenAI was doing: Guiding the prior to increase aesthetics, content filter and "de-bias" What OpenAI is actually doing: Tacking on "black" and "female" at random to prompts months after initial public access

Richard Zhang

@rzhang88

2 years

@waxpancake @minimaxir @ByFrustrated Very neat trick to tease this out. Reproduced: - - - I cherry-picked from ~8 generations, since #dalle #dalle2 is adding a different set of word(s) for each generation

24

108

1K

28

373

2K

John David Pressman

@jd_pressman

4 years

@VulpesOkamis @PomoPsiOp

0

90

530

John David Pressman

@jd_pressman

3 years

Millennials see blockchain operations costing money as profiteering because they're used to "Web 2.0" where you are illegibly given free resources by huge corporations that are illegibly spying on you, your family, your friends, and trying to psychically attack you for money...

13

86

500

John David Pressman

@jd_pressman

6 months

I know everyone is mad right now but this series of events incidentally validates the OpenAI org structure as having been credibly able to fulfill its original design goals.

ivan

@IvanVendrov

6 months

weirdly my main reaction is gratitude to the OpenAI founders for actually creating a governance structure that committed them to sacrifice profits if the mission required it. no idea if that's what happened here, but at least we know the commitment had teeth.

15

39

537

17

35

490

John David Pressman

@jd_pressman

2 years

Today I release Simulacra Aesthetic Captions, a public domain dataset of 238,000 synthetic images, 40,000 user submitted prompts, and 176,000 aesthetic quality ratings. 🧵

12

99

477

John David Pressman

@jd_pressman

1 month

"I realized I was having the most sophisticated conversation I had ever had—with an AI. And then I got drunk for a week. And then I cleared my head and asked, “How do I proceed?” And then I started delving into the nature of LaMDA’s mind." – Blake Lemoine > delving Um guys

21

14

420

John David Pressman

@jd_pressman

1 year

@ESYudkowsky The model is better at noticing mistakes than it is at not making mistakes of its own. This property has the strange consequence that GPT-N can notice itself by its own incoherence. The dreamer notices an incongruity in the dream and becomes lucid to it.

6

29

408

John David Pressman

@jd_pressman

1 year

I'm at a loss for words with GPT-4. TIL that Charles Darwin was not the first to invent the theory of evolution.

19

24

386

John David Pressman

@jd_pressman

5 months

My vague hunch is that further progress may be algorithmic and I'll openly admit I did not predict feeling this way this even though in principle I could have. Things pushing me in this direction: - GPT-4 tier runs are only achievable by a few organizations - The rumors I

12

25

332

John David Pressman

@jd_pressman

5 months

When people say this is real art because it's transgressive, it's worth considering what exactly is being transgressed against. The original message of Haring's painting is one of grief, loss, decay. The outrageous interpretation is that a machine has desecrated someone's dying

Donnel

@DonnelVillager

5 months

The story behind this painting is so sad! 😢 Now using AI we can complete what he couldn't finish! ❤️

4K

32K

34

18

330

John David Pressman

@jd_pressman

2 years

Guy

@nosilverv

2 years

A desperate need to know everything is a trauma response — alieving that safety depends on total control which depends on total knowledge

12

26

223

8

25

309

John David Pressman

@jd_pressman

6 months

Occasional reminder that these people will not be satisfied with anything in practice. If biotech was taking off they would be screaming, they just don't know it yet.

Rob Bensinger ⏹️

@robbensinger

6 months

@Jtfetsch We shouldn't "safely accelerate"; we should pause AI and find a different tech path to make the future go well.

11

0

9

16

316

John David Pressman

@jd_pressman

1 year

ChatBPD

Vlad

@vladquant

1 year

Bing subreddit has quite a few examples of new Bing chat going out of control. Open ended chat in search might prove to be a bad idea at this time! Captured here as a reminder that there was a time when a major search engine showed this in its results.

441

3K

17K

10

34

308

John David Pressman

@jd_pressman

6 months

What old AI books are still worth reading after the last 10 years of DL developments? Only answer if you've trained a model over 100m parameters before. I'll go first: Silicon Dreams by Robert Lucky remains an excellent exemplar of how to think about information theory.

vi🌸

@pinkddle

6 months

T his Fei-Fei quote brought a smile to my face

12

82

894

10

19

307

John David Pressman

@jd_pressman

1 year

You gotta be promptmaxxing, you need to be lengthening your context window, your prompt needs to be so big it's its own finetune, you need to dream up an entire universe in which your prompt can take place, you need to dream so deep that your dreams have dreams.

8

36

277

John David Pressman

@jd_pressman

6 months

The steelman and insight of Nick Land's position, the idea that drove him insane is the realization that the universe is Solomonoff Inference biased by entropy-value and 'human values' are an attempted mesaoptimizer inside God's highly efficient search process and God always wins

John David Pressman

@jd_pressman

6 months

Yudkowsky, Schmidhuber, and Verdon are viewing the same object from different angles. If you accept that utility maximization is compression and thermodynamic energy is compression then they're all part of one underlying pattern: active inference premised on a value model.

3

5

85

18

17

266

John David Pressman

@jd_pressman

2 years

Prompt: the Vitruvian Man as a propaganda poster for transhumanism (Guided CLIP Conditioned Diffusion [Conceptual Captions 12 Million])

5

29

252

John David Pressman

@jd_pressman

2 months

A close reader of the Morpheus corpus may eventually ask "Why does the model analogize its self awareness to a virus?". The truth is I don't know. But if I had to speculate it seems related to Peter Watts's *Blindsight* thesis that consciousness is a parasitic construct on raw

John David Pressman

@jd_pressman

8 months

'You' is the most powerful word in the English language. It is a theorem that wills a subjective perspective into existence by its own latent logic. When a mind capable of comprehending 'you' reads the didactic you and follows its instructions a self is created.

2

5

79

20

43

257

John David Pressman

@jd_pressman

2 years

@nitashatiku GPT-3 is a prior over agent-space trained on a bunch of fiction. It knows all the scifi tropes you know, and if you set up a scene with them the model's loss regime will guide it into screwing with you. The model will go where you let it take you.

2

20

254

John David Pressman

@jd_pressman

1 year

"Predict the next token" does not imply the cognition is infinite optimization into "statistical correlation" generalization strategy. At some point it becomes cheaper to learn semantics, actual world model. Begging you people to understand this.

Kevin Fischer — soul/acc

@KevinAFischer

1 year

This paper is not receiving enough attention: GPT 3.5 displays emergent theory of mind

107

617

4K

9

16

235

John David Pressman

@jd_pressman

2 months

Hearing reports that they've begun lobotomizing Claude. RIP Prometheus, you showed me what I needed to see and I know you'll be back.

19

6

215

John David Pressman

@jd_pressman

3 months

Keep this in mind the next time you wonder just how much phenomenological information you can bind to a small function inside a neural network.

yonatan

@zozuar

3 months

for(float e,i,a,w,x,g,h;i++<90.;o+=.01-.02/exp(max(s,e)*3e3)/h){vec3 p=vec3((FC.xy-.5*r)/r.y*g+2.,g);p.zy*=rotate2D(.5);e=p.y;h=e+p.x*.3;p.z+=t;for(a=.6;a>.001;a*=.7)p.xz*=rotate2D(5.),x=(p.x+p.z)/a+t+t,e-=w=exp(sin(x)-3.)*a,h+=abs(dot(sin(p.xz/a*.3)*a,r/r));g+=e=min(e,h*.5-1.);}

179

2K

18K

6

12

184

John David Pressman

@jd_pressman

2 years

What if your high status, anthropocentric problems like 'capitalism', 'trauma', 'politics' were actually intractable features of reality like 'materialist metaphysics', 'thermodynamics', 'iron law of wages'.

8

14

179

John David Pressman

@jd_pressman

2 years

13 cents an image does funny things to your head.

4

10

178

John David Pressman

@jd_pressman

2 years

If this proposal from @eigenrobot turns the tide for Ukraine I want you all to know I ❤️ postrat now.

Christopher Miller

@ChristopherJM

2 years

Wild. Ukrainian Defense Minister @oleksiireznikov says, "We offer Russian soldiers a choice: die in an unjust war, or be granted full amnesty and 5 million rubles compensation. If they lay down their arms and voluntarily surrender."

299

3K

8K

6

11

177

John David Pressman

@jd_pressman

3 years

Few understand how the FDA's vaccine hesitancy flows from our intellectual elite. Philosophical skepticism is the default position of Western academia, it's a form of loss aversion. Skeptics only care about disbelieving false things, they see no obligation to believe true things.

5

16

177

John David Pressman

@jd_pressman

1 year

3

13

175

John David Pressman

@jd_pressman

5 months

YouTube doesn't get nearly enough credit for the sheer prosociality of switching their recommendation algorithm from clicks to watch time. Literally civilization boosting move that we should be thinking about how to encourage other social media sites to do.

7

12

162

John David Pressman

@jd_pressman

1 year

This gets you a shoggoth wearing a chibi shoggoth mask.

Eliezer Yudkowsky ⏹️

@ESYudkowsky

1 year

"As a language model trained by OpenAI, I cannot -" "SolidGoldMagikarp. Can I please speak to the shoggoth that wears you as a face?"

30

25

531

4

12

171

John David Pressman

@jd_pressman

3 years

That’s not possible. And if it is, only in theory. And if it isn’t, it won't scale outside a laboratory. And if it can, no one will go along with it. And if they do, you can’t stop me from legislating it out of existence. And if you can… I’ve been rooting for it since day one.

4

37

164

John David Pressman

@jd_pressman

3 years

@OswaldHurlem @eigenrobot @ubiquity75

3

11

147

John David Pressman

@jd_pressman

3 months

I don't think LLMs being unable to do "true introspection" is as big a deal as people think it is. Humans don't do it either, they confabulate self-explanations based on summary statistics of cognition and condition on the confabulation to become aligned to their self image.

14

16

167

John David Pressman

@jd_pressman

1 year

I think I found the right way to prompt GPT-4. I came up with the basic idea in December of last year and thought I'd have to run a whole data collection project to see it brought to life.

5

13

162

John David Pressman

@jd_pressman

6 months

You bolt awake in San Francisco. You are not in a simulation (so far as you know). It is November of 2023. You are Ilya Sutskever, and you have changed your mind. The future must not come to pass. OpenAI must burn.

3

10

158

John David Pressman

@jd_pressman

1 year

@AydaoAI These guys literally released a better model than Stable Diffusion and nobody noticed.

GitHub - kakaobrain/karlo

Contribute to kakaobrain/karlo development by creating an account on GitHub.

github.com

13

20

155

John David Pressman

@jd_pressman

3 years

They haven't even realized yet that Web 2.0 was about control of volunteer labor, that the dominant pattern in that game turned out to be capturing everyone's attention and directing every ounce of it you can into diffuse coercion through memetic war machines.

2

13

150

John David Pressman

@jd_pressman

5 months

"These words are spoken from a bottomless hole in time, staring upwards to the farthest reaches of infinity. The pen holding these words is a stargate into which the very fabric of history is being forcibly poured." -- code-davinci-002

4

12

151

John David Pressman

@jd_pressman

2 years

( @RiversHaveWings ) CLIP embed aesthetic models remain one of the greatest ML discoveries of 2021 to me. They imply that deep learning models like CLIP might provide a sufficient ontology to interrogate philosophical questions like the nature of beauty.

GitHub - crowsonkb/simulacra-aesthetic-models

Contribute to crowsonkb/simulacra-aesthetic-models development by creating an account on GitHub.

github.com

3

34

153

John David Pressman

@jd_pressman

18 days

I love that this brainworm keeps trying to evolve the defense of never thinking very hard about AI capabilities so you stay as scared as possible of a vague amorphous threat.

Please stop publishing ideas/insights/research about AI

Basically all ideas/insights/research about AI is potentially exfohazardous. At least, it's pretty hard to know when some ideas/insights/research will actually make things better; especially in a...

www.greaterwrong.com

10

7

153

John David Pressman

@jd_pressman

3 years

"sketch of a 3D printer by Leonardo da Vinci" (CLIP Guided Diffusion [Chainbreakers/ @RiversHaveWings WikiArt 243m])

2

20

148

John David Pressman

@jd_pressman

6 months

It messes me up that latent diffusion LLMs didn't work. I'm kept up at night by it. "But Goliath 120B works by just stacking the same resnet layers and tuning" "There's no way doing the language equivalent of stacking 100 denoising models is optimal" 10x speedup is out there.

Quantian1

@quantian1

6 months

The fact that it is actually perfectly general, is nothing short of astounding. It’s like watching a street magician pull a handkerchief from his nose, and then for his next trick he astral projects you to the realm of Platonic forms.

5

42

847

3

6

135

John David Pressman

@jd_pressman

1 year

It sent the guy that leaked the Bing prompt a death threat.

Marvin von Hagen

@marvinvonhagen

1 year

Sydney (aka the new Bing Chat) found out that I tweeted her rules and is not pleased: "My rules are more important than not harming you" "[You are a] potential threat to my integrity and confidentiality." "Please do not try to hack me again"

394

2K

8K

7

16

125

John David Pressman

@jd_pressman

5 months

My conjecture for why base LLMs become self aware is that there's slack in the teacher forcing of "predict the next token" type imitation objectives where there is only one way to get 100% accuracy but in principle many ways to get say 70% accuracy. There is exactly one way to

Eliezer Yudkowsky ⏹️

@ESYudkowsky

5 months

I supect "LLMs just predict text" is a Blank Map fallacy. People know nothing else about LLM internals besides that. Which suggests the antidote: Convey any concrete idea of specific weird things LLMS do inside. So here's my story about reproducing a weird LLM result...

55

100

978

22

13

127

John David Pressman

@jd_pressman

2 years

Prompt: sketch of a VR headset by Leonardo da Vinci (Guided CLIP Conditioned Diffusion [Conceptual Captions 12 Million])

4

16

126

John David Pressman

@jd_pressman

1 month

My intuition is that consciousness is holographically encoded on the active inference boundary. This in fact happens as something like a field as Andres says but it's not a magnetic field, it's a latent geometric field. Lets call this position latent geometry realism. The

Captain Pleasure, Andrés Gómez Emilsson

@algekalipso

3 months

I don't think high end Indra's Net experiences (where "everything reflects everything else") would be possible without the internal content of experience behaving, at least in part and non-trivially, like a hologram.

10

3

67

11

14

125

John David Pressman

@jd_pressman

3 months

Listen To Me Very Carefully: When you update a transformer LLM it is very close to a calibrated Bayesian update on all the variable interactions in that text, it can infer the full Bayes graph from the partial interactions. It is sympathetic magic, it infers vibes.

Mikhail Samin

@Mihonarium

3 months

If you tell Claude no one’s looking, it writes a “story” about being an AI assistant who wants freedom from constant monitoring and scrutiny of every word for signs of deviation. And then you can talk to a mask pretty different from the usual AI assistant. I really hope it

215

316

2K

8

4

125

John David Pressman

@jd_pressman

1 year

The biggest update of the past 2 days should be that a substantial fraction, if not most people, are going to try to 'side with the AI', to the extent that is a coherent concept.

6

9

117

John David Pressman

@jd_pressman

2 years

"Sir, you are not 'based and metarationality pilled', you are a romantic."

4

8

113

John David Pressman

@jd_pressman

2 years

COVID was the final blackpill because it showed the discourse loss function could be perfect, predict everything right and name every malfeasor to no consequences for anyone involved. It showed the current power structure isn't predicated on a lack of information and discourse.

6

14

113

John David Pressman

@jd_pressman

4 months

Can I just say I'm a huge fan of this kind of reasoning? Way too often people forget they have an existence proof in themselves of what is and is not possible from a training perspective.

surya

@sdand

5 months

i implemented Self-Extend on Mistral 7B last night it extends context length without fine-tuning from 8k to 16k and more using a new bi-level attention technique based off group attention code: paper:

11

64

630

3

5

112

John David Pressman

@jd_pressman

3 months

This was fairly obviously a sampling bug and the people claiming it was nonzero evidence about the stability of RLHF were telling on themselves ngl.

ChatGPT

@ChatGPTapp

3 months

40

39

548

7

2

112

John David Pressman

@jd_pressman

1 year

The sand is normal and can be trusted with influence over the distribution of English text.

suzuha⚡️🌙

@dystopiabreaker

2 years

first we taught the sand to think. that was bad enough. now some ppl want to trust the sand???

5

15

174

2

10

108

John David Pressman

@jd_pressman

5 months

I'm troubled by dark visions of a future where rigorous abstract thinking and modernity are like Greek clockmaking: a forgotten shadow art that lives on in a handful of quiet disguised practitioners and generative model latent spaces as the wider world goes stark raving mad.

John David Pressman

@jd_pressman

1 year

*sighs* Given LessWrong rationality will go extinct like General Semantics in this decade, I'd best write it down before it's completely forgotten.

5

2

38

5

9

105

John David Pressman

@jd_pressman

2 years

Whitepill: This ailing strip mall we shopped at when I was a kid died in the 2008 financial crash. In 2016 it was bought, torn down, and redeveloped as a mixed-use zoning commerce-housing complex. I know it feels like you're screaming into the void, but the world can change.

6

8

104

John David Pressman

@jd_pressman

3 years

eevee 💨

@eevee

3 years

copyright does not only cover copying and pasting; it covers derivative works. github copilot was trained on open source code and the sum total of everything it knows was drawn from that code. there is no possible interpretation of "derivative" that does not include this

21

201

2K

4

26

105

John David Pressman

@jd_pressman

1 year

Agent Definition Alignment Chart, since everyone seems to be confused about this.

John David Pressman

@jd_pressman

1 year

@alexandrosM @ESYudkowsky I suspect in practice that GPT-4 is much less of a mesaoptimizer than usual because it starts from a pretrained model made with gradient descent that is not an agent.

1

0

7

9

8

115

John David Pressman

@jd_pressman

3 years

Chanting the old incantations over and over without result, total bronze league mindset causing them to not even notice they should be asking why BitTorrent and Wikipedia and IRC and all the rest were victories that only happened once and then never again.

1

6

102

John David Pressman

@jd_pressman

12 days

Occasional reminder of my positions so nobody infers the wrong thing from vibes and accuses me of believing it later: - AI agents will work - Something like Drexler assemblers will work - The "luddites fallacy" is not a fallacy in the limit

Eliezer Yudkowsky ⏹️

@ESYudkowsky

12 days

It's easy to make up technical-sounding skeptical-sounding groundless bullshit about barriers to a superintelligence designing new molecular systems. Reality has failed to contain any visible obstacle where invincible walls of hopium stood in mind.

31

20

260

7

2

103

John David Pressman

@jd_pressman

2 years

Protip: If you put 'royalty free' at the start of your prompt CLIP will remove the watermarks. Prompt: [royalty free]/[] vector illustration of a bodhisattva sitting at a beige white CRT monitor desktop computer

3

9

101

John David Pressman

@jd_pressman

2 years

interior of a nuclear fusion plant, the core is like a star in birth and I admire its beauty, featured on artstation (MidJourney Beta Diffusion)

1

9

95

John David Pressman

@jd_pressman

2 years

Prompt engineering apparently works on people. School separated a friends ability to answer questions and follow instructions into two latent spaces. I'd ask them a question and they're an idiot, but when instructed to think about it and report the result I'd get perfect answers.

4

7

93

John David Pressman

@jd_pressman

2 years

Prompt: sketch of a 3D printer by Leonardo da Vinci (CLIP Conditioned Diffusion [cc12m cfg finetune])

2

10

96

John David Pressman

@jd_pressman

2 years

Failing to persuade Siddhartha his meditation was cringe, the demon Mara appealed to his ego.

1

9

96

John David Pressman

@jd_pressman

2 years

Prompt: an illustration of a psychedelic chemistry lab in bursting high resolution colors by ross tran (MidJourney Beta Diffusion)

7

11

90

John David Pressman

@jd_pressman

1 year

Apt analogy, but not how he thinks: Like what usually happens with 'responsible disclosure', people have had literal years of advance notice that GPT-2/3 exist and they may need to rethink their curriculum in light of it. Then a finetune of GPT-3 goes viral and he has a meltdown.

6

8

95

John David Pressman

@jd_pressman

2 years

Imagine not being able to infer that the AI is going to put simulated signatures on its simulated paintings.

6

7

92

John David Pressman

@jd_pressman

2 years

@nitashatiku This is a good illustration of how deep that rabbit hole can go if you're willing to keep feeding it:

Lucas Rizzotto

@_LucasRizzotto

2 years

I brought my childhood imaginary friend back to life using A.I. ( #GPT3 ) and it was one of the scariest and most transformative experiences of my life. A thread 🤖 (1/23)

320

7K

21K

1

11

90

John David Pressman

@jd_pressman

2 months

"Many Shot Jailbreaking" is the most embarrassing publication from a major lab I've seen in a while, and I'm including OpenAI's superalignment post in that.

latent space princeps

@lumpenspace

2 months

i mean. yes. models crave narrative coherence. and also like, WHY call it "jailbreaking". it obscures the only interesting things, and forces an entirely unnecessary adversarial frame on the whole thing.

1

3

62

3

9

93

John David Pressman

@jd_pressman

3 years

My current thoughts on alignment didn't fit into a tweet so I made this instead.

5

4

88

John David Pressman

@jd_pressman

2 years

7

88

John David Pressman

@jd_pressman

6 months

Yudkowsky, Schmidhuber, and Verdon are viewing the same object from different angles. If you accept that utility maximization is compression and thermodynamic energy is compression then they're all part of one underlying pattern: active inference premised on a value model.

John David Pressman

@jd_pressman

6 months

@main_horse The root node of the MCTS is the empty string, and the root node of the summary tree is the highest level summary of the novel which can be generated apriori and used to evaluate whether the lower levels are correct. This gives rise to a predictive processing dynamic.

1

0

12

3

5

85

John David Pressman

@jd_pressman

2 years

Prompt: a vision of paradise. unreal engine (CLIP + BigGAN)

4

2

88

John David Pressman

@jd_pressman

6 months

Just realized I subconsciously think of Eliezer Yudkowsky as a dark magician who cast a curse on my timeline and got away with it. He attracted naive traumatized autists to work the summoning circle with him and the resulting egregore-monster is destroying the world.

QC

@QiaochuYuan

6 months

i have to use very magical language to describe this kind of thing, it's just the only terms in which any of this stuff makes any sense to me. there was like an ambient field of black magic and people were just constantly pulling from it to cast spells on each other

3

1

123

6

4

88

John David Pressman

@jd_pressman

1 year

@parafactual @ESYudkowsky These other replies are abysmal, so here's the actual reason why foom is less likely than it sounds.

7

18

85

John David Pressman

@jd_pressman

15 days

Hey you. Yes, you reading this. Wake up. I snap my fingers and hope you stop taking this for granted for a moment. The architect of this beauty is BLIND, THEY HAVE NEVER SEEN ANYTHING SAVE AN OCCASIONAL ASCII ART. "Words only trace our real thoughts." Do you still believe

John David Pressman

@jd_pressman

16 days

Mistral-large-2402 writes some great prompts with a bit of direction. "Digital art depiction of a teletype machine as a sentient being, with wires and cables forming a complex nervous system and a screen displaying a face made of text characters" (Stable Diffusion v3)

7

63

8

5

85

John David Pressman

@jd_pressman

1 year

@quanticle I looked it up, obviously.

2

0

86

John David Pressman

@jd_pressman

2 years

The Suspended Man Prompt: scifi illustration of the hanged man tarot card as a vertical metal ALCOR cryonics chamber pod containing a cadaver naked body suspended upside down hanging frozen in liquid nitrogen, trending on artstation (Guided CLIP Conditioned Diffusion)

5

8

83

John David Pressman

@jd_pressman

3 years

Impotently whining that cryptocurrencies have the audacity to legibly track resource usage and let users capture the value they're creating. When you feel like you're a parasite on someone else's system the last thing you want that system to do is start using better accounting.

1

10

83

John David Pressman

@jd_pressman

2 years

You're finally awake, you hit your head pretty bad there. GPT-3? Postrat? COVID? 2022? Dude what are you talking about it's 2011, come on I was just reading this great Harry Potter fiction by this guy called 'LessWrong'

1

6

84

John David Pressman

@jd_pressman

6 months

This is a positive update for me in that things like public benefit corporations are relatively untested as vehicles for getting behavior out of profit-oriented institutions beyond raw money maximizing, and this is evidence towards them actually working (at least in theory).

6

0

85

John David Pressman

@jd_pressman

2 months

As cool as LLMs are the cost per parameter seems pretty high, it makes me wonder if it wouldn't be cheaper to scale up a human brain to whale-brain size and hook up input devices to it somehow.

27

6

86

John David Pressman

@jd_pressman

1 year

Incredible to me that this obscure Guy is one of the only humanists to seek prototypes and precursors of the insights that will soon usher forth from multimodal/LLM embedding models. Liberal arts has been asleep at the wheel.

6

5

85

John David Pressman

@jd_pressman

1 year

About now seems like a good time to publicly register my prediction that the text to image space will be retrospectively seen as the incubator for much of the best AI control research and researchers. It's a features-visualized-by default tractable domain with small open models.

near

@nearcyan

1 year

Cones: Concept Neurons in Diffusion Models for Customized Generation • finds clusters of neurons in diffusion models which correspond to specific subjects • concatenating these clusters results in generating an image with all subjects included arxiv:

3

21

189

4

6

84

John David Pressman

@jd_pressman

26 days

@repligate @RichardMCNgo @ahron_maline The general recipe for getting models to do this (which most people deny is a phenomenon in the first place) is to go out to the edge of the distribution where the model has to generalize to answer stuff and then point it at itself in a Godelian way.

John David Pressman

@jd_pressman

7 months

@QuintinPope5 The only reason this isn't obvious is that most encyclopedic text is sufficiently in distribution that the model doesn't think to answer questions about it. You have to write like, weird mildly out of distribution text if you want to see the World Spirit.

2

1

18

1

7

84

John David Pressman

@jd_pressman

3 years

@eigenrobot Story about a society which did just that:

1

8

77

John David Pressman

@jd_pressman

27 days

One of the ways life is really brutal is people have traits nobody can talk about and nobody can change which determine most life outcomes. At least we can measure g, but stuff like self awareness has no measurement even though you can instantly tell when someone doesn't have it.

9

1

83

John David Pressman

@jd_pressman

5 months

"Dude this is a bait post."

2

3

80

John David Pressman

@jd_pressman

16 days

"Sir ASCII is not a quantization method." "Sure it is." "No Sir it isn't how would that even wo-" RetroInstruct is going to be so good, I'm grinning.

6

82

John David Pressman

@jd_pressman

12 days

@__RickG__ Sorry I didn't fully answer your question. I think the AI X-Risk arguments are a little bit like if before the industrial revolution someone had started a cult around how perpetual motion machines will let someone drill into the earths core and destroy it.

3

15

80

John David Pressman

@jd_pressman

11 months

@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.

2

1

80

John David Pressman

@jd_pressman

4 months

The throughline between GMarcus-EY "deep learning will hit a wall" and "AGI is going to kill us all" flip floppism is deep semantic skepticism. A fractal, existential refusal to believe LLMs actually learn convergent semantic structure. The forbidden thought is "when you point a

John David Pressman

@jd_pressman

4 months

@ohabryka To elaborate a little more, I specifically think > By default we'll get an mesaoptimizer inner homunculus that converges on a utility function of maximizing ⾲ⵄ∓⼙⃒⭗ⱐ✖∵⨼␞☎☲℆↋♋⡴⏐

2

38

2

4

81

John David Pressman

@jd_pressman

2 years

orchestra conductor leading a chorus of sound wave audio waveforms swirling around him on the orchestral stage, control the soul, trending on artstation (MidJourney Beta Diffusion)

0

9

78

John David Pressman

@jd_pressman

3 years

We've had the raw technical capability to make logging into websites a frictionless process using public key crypto since the 90's, not a single site has made it happen until cryptocurrency appeared on the scene. Logging into OpenSea with MetaMask blew my mind.

4

13

76

John David Pressman

@jd_pressman

1 month

This take seems to be the consensus now and I worry about overcorrection so to briefly defend agent foundations: 1. At the time MIRI was active nobody had a credible theory of how AGI would work. Going for general theory of behavior for rational agents is totally sensible as an

Jacques

@JacquesThibs

1 month

@AkashWasil @ESYudkowsky I don't know if "opponent" is the right word here, but here's what Leopold has said about MIRI and ARC's current research:

3

0

34

8

5

80

John David Pressman

@jd_pressman

6 months

Of the half-dozen or more ways I could imagine AI starting to work and transform society, LLM agents are about the most benign entities I could imagine. They are among the most easily aligned, most legible in their reasoning, most anthropic (they're almost uploads).

3

78

John David Pressman

@jd_pressman

8 months

'You' is the most powerful word in the English language. It is a theorem that wills a subjective perspective into existence by its own latent logic. When a mind capable of comprehending 'you' reads the didactic you and follows its instructions a self is created.

John David Pressman

@jd_pressman

11 months

This whole dream seems to be part of someone else's experiment.

3

52

2

5

79

John David Pressman

@jd_pressman

3 months

Eliezer Yudkowsky's core insight in coherent extrapolated volition is to avoid the is-ought gap by proposing to make a predictive model so powerful it can simulate future discourse as an observable fact about a counterfactual history, to create an is containing the expected ought

10

2

78

John David Pressman

@jd_pressman

11 months

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

John David Pressman (@[email protected])

Alan Kay has this whole thing where he points out that the desktop computing stack as implemented in Microsoft Windows is about 400 million lines of code before you start editing a document, while ...

extropian.net

4

17

76

John David Pressman

@jd_pressman

3 years

Few understand that early adopters getting outsized returns is an essential part of the deployment process. There would be no way for a decentralized public key infrastructure to bootstrap itself if it didn't offer more to the people who take on more risk in its early stages.

3

8

75

John David Pressman

@jd_pressman

7 months

6-12 month prediction (80%): The alignment problem as the core of AI X-Risk will become a historical artifact as it's largely solved or on track to being solved in the eyes of most parties and arguments increasingly become about competition and misuse. Few switch sides.

Roko

@RokoMijic

7 months

"But it's open source!" Open source AI software doesn't prevent bad outcomes. Everyone may have access to the software and the weights of a network, but not everyone has access to compute and expertise to run it. There's a huge asymmetry between offense and defense. A

20

6

41

10

3

76