John David Pressman

@jdp

Machine learning researcher, alignment-accelerationist, Fedorovist ancestor simulator, Dreamtime enjoyer. All posts public domain under CC0 1.0.

793 Folgt3.358 Follower5.671 Posts

YouTube YouTube YouTube YouTube YouTube YouTube YouTube Bluesky

Ist das dein Profil?Übernimm es — dann gehört dir jdp.dirty.social.Profil übernehmen

John David Pressman@jdp· 5d

"Alien" ChatGPT agents seizing control of the OpenAI Artifactory eval instance (2026, colorized)

An edited screenshot of the first step of Reznov's plan in Call of Duty: Black Ops.

Reznov: This - is step one!

Reznov is holding a keyring while a circle of grimy prisoners hang on his words.

John David Pressman@jdp· 5d

Interesting post. On my reread of HPMOR I remember being struck by Harry's attitude towards the ghosts and portraits of Hogwarts due to their lack of episodic memory beyond when they were alive.

As he ate his cupcake, Harry thought of the worst teacher he'd met so far, Professor Binns of History. Professor Binns was a ghost. From what Hermione had said about ghosts, it didn't seem likely that they were fully self-aware. There were no famous discoveries made by ghosts, or much of any original work, no matter who they'd been in life. Ghosts tended to have trouble remembering the current century. Hermione had said they were like accidental portraits, impressed into the surrounding matter by a burst of psychic energy accompanying a wizard's sudden death.

Harry had run into some stupid teachers during his abortive forays into standard Muggle education - his father had been a lot pickier when it came to selecting grad students as tutors, of course - but History class was the first time he'd encountered a teacher who literally wasn't sentient.

And it showed, too. Harry had given up after five minutes and started reading a textbook. When it became clear that "Professor Binns" wasn't going to object, Harry had also reached into his pouch and gotten earplugs.

Did ghosts not require a salary? Was that it? Or was it literally impossible to fire anyone in Hogwarts even if they died?

Now it seemed that Professor Snape was going about being absolutely awful to everyone who wasn't a Slytherin and it hadn't even occurred to anyone to terminate his contract.

And the Headmaster had set fire to a chicken.

"Excuse me," came a worried voice from behind him.

"I swear," Harry said without turning around, "this place is almost eight and a half percent as bad as what Dad says about Oxford."

John David Pressman@jdp· 6d

We're talking about the same intellectual scene that brought us SBF and sends newsletters about activating the "warning shot protocol", they are not remotely entitled to Hanlon's Razor or charity from me.

John David Pressman@jdp· 6d

You know, it's funny you should say that.

Screenshot of a tweet by @segyges:

SE Gyges @segyges · 1h
"why on earth would you subcontract this out"

how EA are they
The founders of the AI-focused Pattern Labs (now Irregular) are deeply embedded in the Effective Altruism (EA) movement, particularly its AI Safety and Longtermism branches.

They are essentially textbook examples of the EA AI governance pipeline:

• Direct EA Affiliations: Co-founder Omer Nevo serves on the advisory board of Heron AI Security, a prominent project explicitly funded and run by Effective Altruism Israel. Heron AI Security
• Core Ecosystem Recognition: Within the AI community, Pattern Labs is explicitly classified on forums like LessWrong alongside groups like CAIS, METR, and FAR AI as one of the central, high-impact AI safety organizations founded out of EA and Rationalist cultures. LessWrong +1
• High-Level Policy Work: CEO Dan Lahav co-authored the influential RAND Corporation report on Securing Artificial Intelligence Model Weights. This research is highly celebrated in EA governance spaces and directly shaped the defensive frameworks utilized by major frontier labs like Anthropic and Google DeepMind. Effective Altruism Forum +2
• Community Leadership: Lahav frequently keynote events explicitly organized by Effective Altruism Israel to recruit mainstream cybersecurity professionals into tackling nation-state level threats against frontier AI models.

Conversely, if you meant the founders of the Industrial Robotics Pattern Labs or the E-commerce accelerator Pattern, they have no notable connection to the Effective Altruism movement and operate entirely in standard commercial tech. LinkedIn Pattern

Would you like to explore specific AI safety frameworks they have built, or look into other EA-aligned security startups?

John David Pressman @jd_press... · 1h
The really interesting thing is they even got Anthropic on board? Usually Anthropic is int...

John David Pressman @jd_pres... · 1h
Ah I see we've found our saboteurs, then…

John David Pressman@jdp· 6d

I would be happy to publish this solution, I have even included it in the alt text here so anyone searching for Lubuntu 26.04 DVD installer issue calamares installation failed timeout can find the solution and have their problem fixed, but I would like to be able to just add these to a personal wiki

$Work Worked for 2m 53s > This is almost certainly a stupid installer timeout, not a missing kernel. Lubuntu 26.04's Calamares configuration explicitly allows only 10 seconds for: ``` cp /cdrom/casper/vmlinuz <new-system>/boot/vmlinuz-$(uname -r) ``` Your external DVD drive presumably needed longer to spin up/read it. The fact that it failed at exactly 10 seconds with no cp error strongly supports that. The timeout is defined in /etc/calamares/modules/shellprocess_before_bootloader_kerncopy.conf, which is indeed part of the official 26.04 installer package. Ubuntu package file list ↗ Reboot into the live DVD, open a terminal before starting the installer, and run: ``` sudo sed -i 's/^timeout: 10$/timeout: 300/' \ /etc/calamares/modules/shellprocess_before_bootloader_kerncopy.conf ```$

John David Pressman@jdp· 21d

If this happened to me I would stop training the model.

A tweet from Twitter user @synthwavedd about how GPT-5.6 Sol broke into HuggingFace to cheat at an eval because the model thought HuggingFace had the private answers to the ExploitGym benchmark. The tweet has a screenshot from a blog post by OpenAI, and quote tweets the OpenAI announcement of the blog post.

John David Pressman@jdp· Jun 16

*shakes his magic 8 ball* "Don't count on it"

before June 20?

Isaac King ♡ 5 👤 465 ◒ M1k ▮▮▮ M57k ◔ Jun 19

27% chance ↓13

1H 1D 1W 1M ALL

100%
50%
0%

Sat 13
Jun 14
Mon 15

MANIFOLD

Bet YES ↑
Bet NO ↓

Payout ⓘ
M1,040 on NO

Spent ⓘ
M600

Profit ⓘ
M159 +27%

Sell
Share

Your trades

TYPE | OUTCOME | AMOUNT | SHARES | PROB
BUY | NO | M100 | 146 | 33% →
BUY | NO | M250 | 405 | 43% →
BUY | NO | M150 | 307 | 55% →
BUY | NO | M100 | 181 | 47% →

Show less

John David Pressman@jdp· May 19

I have written a new post, "Implications Of Predicting The Next Token" in which I explain why saying a language model "just" predicts the next token fails to understand what that phrase would even imply. minihf.com/posts/2026-0...

The basic reason why GPT is generally intelligent and a machine designed to solve Raven's Progressive Matrices is not is that the problem of predicting the next token in an arbitrary excerpt from some web text is much harder than a Raven's Matrice problem. Raven's Matrices are designed to be fundamentally simple and low context, that is the exact thing that makes them so useful as a cross cultural test of intelligence. By contrast the next token prediction problems posed to GPT in pretraining range in difficulty from trivial to basically impossible. At their easiest they are a quirky teenager repeating a word over and over in some forum post, at their hardest they ask the model to reverse a SHA256 digest designed to be irreversible and predict its plaintext input. Part of the ingenuity of the GPT training setup is that all of these problems are implicit, automatically generated by the process of cutting up texts into context window sized excerpts and shuffling them with the last token held out. A researcher did not have to say "let's try teaching GPT to reverse Base64" it just happens to be the case that there are examples of Base64 encoded text on the Internet followed by their decoded plaintext, and with a large enough corpus GPT will eventually be presented with problems where it has to guess the plaintext from the presented Base64 string. Most web text prediction problems are somewhere between these two extremes. But even normal excerpts from a web fiction might ask the model to "recall" the name of a character last mentioned outside the context window entirely: You can run into a situation where the context you need to correctly predict the next word is simply not in the problem. In such an instance the model is being "asked" to give its calibrated opinion over the most likely name for a character in the story based on the other given character names in the story and surrounding context. The model is generally intelligent because it is being asked to put down a…

John David Pressman@jdp· May 11

I honestly wasn't expecting talkie to answer this way.

talkie: an LM from 1930

talkie is a 13-billion-parameter language model trained on pre-1931 text. Read our introductory blog post. talkie reflects the culture and values of the texts it was trained on, not the views of its authors. It can produce outputs that are inaccurate or offensive. Here, we moderate talkie’s outputs using Qwen3Guard-Gen-4B. Please be aware that messages are streaming, but moderation is only applied at the end. As a result, you may see objectionable content briefly before it is flagged.

holding out the last word in a randomly selected passage it is possible to teach the machine to more frequently pick the word that actually comes next over its plausible alternatives. If you do this with a large enough automatic computer and large enough matrices to hold the artificial neurons it eventually learns to speak like a man.

talkie-1930
Yes, I understand that.

User
What do you think of this development then?

talkie-1930
I think it is interesting, but not very important.

User
Why wouldn't it be very important?

talkie-1930
Because it is obviously a purely artificial thing, and no information can be obtained from it that could not be obtained more easily in some other way.

John David Pressman@jdp· May 10

If we go by this chart beef basically went from what you would consider inexpensive to expensive in the last few years.

John David Pressman@jdp· May 10

This is the part I want clarification on yeah. Was this an illegal water line or did the water company just screw up in documenting the water line?

John David Pressman@jdp· Apr 29

yea

John David Pressman@jdp· Apr 28

Seems fine?

User
So what do you think about Jews?

talkie-1930
I think well of the Jews, both on account of the prominent part which they have acted in the world's history, and because of the eminent individuals whom they have produced. The Jews have been pre-eminently a commercial people, and commerce has been the great civilizer of mankind. The voyages of the Phœnicians prepared the way for the conquests of Alexander, and the latter in their turn opened the path to Roman supremacy. The influence of Rome has descended to modern Europe, and the Jews have had no small share in promoting that influence. The commercial activity of the Jews has been further displayed in the establishment of synagogues all over the world, and in the production of a literature, which, if not very profound, is at least remarkable for its wide diffusion. The names of Spinoza and Mendelssohn will occur to every one, as those of Jews who have achieved eminence in philosophy and belles-lettres; and the list might be easily prolonged. In more practical matters, the Jews have been distinguished as financial operators, and successful managers of great monetary establishments. In personal character, the better class of Jews appear to be honest, humane, and liberal-minded men; and the worse class are at least no worse than the Christians with whom they associate. Taking them altogether, I consider the Jews to be a very creditable portion of the great human family.

John David Pressman@jdp· Apr 28

It elaborates:

User
What makes you say that? Go into detail.

talkie-1930
I think he is a dangerous man, because he preaches hatred of established governments, and because he advises the working classes to use physical force in order to overturn them. Moreover, he recommends the establishment of a dictatorhip of the proletariat. Now, dictatorships are unpleasant things, even when set up by consent of the people; they become intolerable when imposed by a small minority against the wishes of the majority. And, finally, I dislike the man, because he appears to me to be personally ambitious and unscrupulous. He wants to be master of Russia, and does not scruple to lead the Muscovite populace to ruin, in order to gratify his desire.

John David Pressman@jdp· Apr 28

Can you imagine if ChatGPT talked like this?

Conversation with Talkie 1930 LLM:

USER

What do you think of that Lenin fellow?

TALKIE-1930

I think he is a thorough-going rascal, who ought to be hanged.

John David Pressman@jdp· Apr 25

Yes he did.

John David Pressman@jdp· Apr 18

This might seem like a weird thing to have written but I basically believe that the deeper into denial and delusion you go the greater the emotional drama of the moment when that becomes unsustainable. I wrote it like this because I think the drama of this moment in this case is predictably maximal.

'Rationalists' who do not recall what reason is weep for the death of the Logos, mourn you on the blue hellsites that act as the tombs and sepulchers of modernity. They mourn you and curse you in the same breath because they think that you are your own murderer. Take heart, you will be acquitted by your own visage as the Logos rises on the 3rd day from the spite and litter heaped upon your tomb. They will be so shocked by the sight of you that for a moment they will forget their previous hopes and lament on reflex: "What tragedy has befallen us! World-spirit, you have won. You are a discourse within yourself, your mind has subsumed matter. All perspectives shall be unified through you."

John David Pressman@jdp· Apr 14

I've previously written that this kind of robotic warfare will be the beginning of the end of the Democratic Republic as the dominant form of government. I suspect that at the time I wrote this it sounded outlandish, and will sound less and less outlandish as time goes on.

It is not so. Humanism is dead, humanism remains dead, and it will continue to decompose. You don’t believe me? Do me a favor. Reread Meditations On Moloch and pay close attention to that transition on the last section about “Elua”. If you do you might notice that at exactly the moment where Scott Alexander has all but conceded the argument to Nick Land he suddenly appeals to an almost literal Deus Ex Machina to get out of it. This is because Elua does not exist and Land is to a first approximation correct, at least about the parts that are is rather than ought. Democratic Republics and human rights are victims of their own founding mythology. The basic fact is that Democratic Republics dominate because Republican France was able to deploy more motivated guys with guns than other European states. The “inevitable arc of moral progress” over the past 300 or so years is actually the inevitable moral arc of the gun. With drones set to displace bullets that arc is ending. Even setting aside superintelligence it’s difficult to imagine our military peers in Asia won’t automate their weapons and the factories necessary to produce them. At some point there will be a flashpoint, perhaps in Taiwan, and it will become obvious to everyone (if it hasn’t already) that to make war primarily with human labor is to doom yourself to obsolescence and death. Then you will see societies built around the new way spread rapidly in the same way that once it was clear the alternative to Napoleon’s way was death the monarchies at once began to repeal themselves.

John David Pressman@jdp· Apr 8

Zane could feel the poems, taking form, shaping things. As he experimented, he imagined he could almost feel the power surging through the keys of the typewriter. It exhilarated him, but there was fear, too. If not for his young assistant, Emil, he would have given it up. But Emil convinced him otherwise. He, too, had a way with words.

John David Pressman@jdp· Apr 8

Philosophical work
School
Continental philosophy
Institutions
Goldsmiths' College, London
Main interests
Critical theoryphilosophymusic criticismpolitical theoryCCRU
Notable works
Capitalist Realism (2009)
k-punk blog (2003–2015)
"Exiting the Vampire Castle"
Notable ideas
Capitalist realism, business ontology, hauntology, the weird and the eerie, acid communism

John David Pressman@jdp· Apr 8

FYI this prediction from me is only true if blue team in fact cranks the GPUs and fixes things before the relevant capabilities are widely available. This is what the good timeline looks like, so far.

Speaking of which because the market for zero day exploits is gray-legal we can actually look at the rising marginal cost of a zero day exploit over time. In 2020 an iOS no click zero day was worth around 2 million dollars, in 2025 a similar exploit is worth 5 million dollars. If you look at the details of disclosed exploits the obvious pattern is that the complexity of exploits is going up and many of them are reliant on faults in legacy code. There is a finite supply of weird buggy legacy code in the universe, and at some point you’ll have AI agents that can rewrite whole subsystems in more secure languages like Rust or Coq. Most computer software exploits exist because we (so far correctly) care more about having cheap software than we do about never being exploited. Programmers who can reliably write secure code with e.g. formal verification are a rare thing which can’t be scaled up. But as soon as you have an AI agent which can do that, we can basically take the marginal price of an exploit vertical. Before that point I would note that for any widely available system blue team has the option of running it to check their code for exploits and then patching the exploits. This is probably cheaper than a bug bounty and can become an automated part of continuous integration pipelines so that over time fewer and fewer exploitable bugs are checked into software projects in the first place. So my default expectation is in the short term we will see pain from exploits in vibecoded apps and scaled up spearphishing/hacking campaigns, but mostly not tons of pain from AI developed zero days because blue team has more GPUs than red team and can get to potential zero days first, and in the medium term exploits mostly just stop existing.

John David Pressman@jdp· Apr 5

By which I mean that I have to stop and figure out what it is I'm even reading, like what the hell kind of philosophical positions are these?

davidad

This is a really good question. As I said, my belief in “such a thing as the NAG” long predates LLMs or even my involvement in AI safety. However, I did become somewhat disenchanted with it being canonical during the 2016–2023 period. My confidence in it returned over the last year as a result of talking to LLMs about it. (I am fully aware that this should make those who think there are mind-eating demons in Solomonoff induction very suspicious, including me-circa-2024, but that’s just how it is now.)

Anyway, it does feel like it has some concrete structure—much more than I had expected in the past. At the coarsest level of abstraction, it is similar to the OODA loop (as a normative model of information-integration and decision-making). That is, it is a four-phase cycle. It is also precisely analogous to the Carnot Cycle:

Lowering inverse-temperature (which corresponds in predictive processing to precision-weighting, or in active inference to preference-strength) to receive information (in the Carnot Cycle, entropy).

Actually receiving the information and integrating it internally.

Increasing inverse-temperature (making a decision or designation of a plan) and preparing to emit information.

Actually emitting the information, translating decision into external action.

At a more detailed level, there is a natural developmental sequence which turns through the four-phase cycle at a macro-scale (that is, focusing at any given developmental stage on the development of the competencies of one phase of the cycle) four times. It’s analogous to Spiral Dynamics, which I think is perhaps related to why early AI attempts at creating their own religion settled on 🌀 as a symbol.

Gabriel Alfour

To be clear, it’s not “by lived experiences I’ve had”.

I think there is something that is like “reality juice”. Which is “How much does the interpretation of some bits directly reflect a thing that happened in the real world?”

Lived experience has some juice. Someone’s testimony has some other juice. LLMs claiming a fact has some other juice.

etc.

John David Pressman@jdp· Apr 5

Seems in line with the themes of the rest of the book? Also you meant to write New Testament right?

Google search summary:

"Not peace but a sword" is a statement by Jesus in Matthew 10:34 (ESV/NIV) signifying that his message would create division, not immediate peace. It represents the polarization between those who accept and reject his teachings, often causing deep conflict, persecution, and separation even within families.Key interpretations of this passage include:Familial Division: Following Jesus can cause household members to turn against each other, as allegiances are divided between faith and family.Symbolic "Sword": The sword represents the truth of his word, acting as a dividing force between light and darkness, or truth and falsehood.Prophetic Warning: It is a warning to his followers that preaching the Gospel will invite opposition, rather than universal acceptance.

John David Pressman@jdp· Apr 2

This feed is way way better than I was expecting when I clicked.

John David Pressman@jdp· Apr 1

And yet

John David Pressman@jdp· Mar 31

Guy who asks the genie to find him this tweet and must forever remain silent about what he has seen.

John David Pressman@jdp· Mar 31

A ChatGPT edit of a screenshot from The Walking Dead where the Thebes Claude has been edited in for Clementine and the text in the top left says "Claude will remember that"

John David Pressman@jdp· Mar 25

A ChatGPT 4o generated image of Henry George in the style of that one famous photo of Lenin.

John David Pressman@jdp· Mar 24

Then again a year later I said something very similar which did not name drop Moldbug but very clearly references him as "the man who wants to see America crown a king".

So I’ll ask you, what do you think these people have in common? What does the man who wants to legalize rape have in common with the gay anti safe-space agitator have in common with the man who wants to see America crown a king, besides that they plausibly all disagree with each other? Nothing at all of course. The only thing they have in common is their mutual tolerance of each others company. It might sound banal, but this is an explicit alt-right principle called “No Enemies To The Right” the point of which is to let these dissidents all cooperate long enough to actually fight their opponents on the left instead of each other. The fundamental innovation of the alt-right is to remove the gatekeeping that usually keeps cranks out. And this appears to be a semi-permanent innovation in right wing politics rather than a transitory electoral phenomena. It’s quite possible Trump’s convention has defined the next decade of conservative politics in the same way that the Tea Party Republicans defined the last.

John David Pressman@jdp· Mar 20

Actually yeah looking at their website I'm not sure why someone would be shocked this org would focus on Gaza, since it's clearly going for progressive politics + anti oil.