I’ve had the idea before of making models that are specialized into specific fields. Currently most big models are Mixture-of-Expert (MoE), where the neurons are separated to make experts inside the model. So you can have the coding expert, math expert etc. I wonder how much of a hack it is and if we won’t find something better soon. But the idea is similar - you decide what experts your model will consist of, and then train the ‘experts’ on expert material in their field.

The problem with the MoE approach is if you get the coding expert when you wanted to ask a linguistic question, it will start talking about data points and running tests. As far as I know the experts are completely separate and there is no crossover, i.e. no neuron that can be used by 2 experts, but maybe this is changing too. However at each step of the generation you may get the input sent through a different expert. From what I understand.

Like I would love a digital design expert that could look at your interface and critique it like an expert designer - graphic, visual, UX, whatever. It’s design at the end of the day. Write its own tests and proofs too if needed. It’s easy enough to mathematically place a grid on a picture, and then use python tools to verify if every item aligns in the grid - if the model doesn’t have vision.

What we are seeing in agentic interfaces though is sub-agents, and the ‘parent’ LLM, the one that you talk to in the session, becomes an orchestrator that spawns and directs the sub-agents (giving them a prompt and clear task, then getting a result back). There have been ideas, from there, to have the orchestrator call smaller models as needed. Then those smaller models could be individual experts, and you could have them on your computer - they just get loaded and unloaded from ram as they are called.

As far as I know though this doesn’t really exist yet though, and there are a few bottlenecks I can think of to work through, but I can definitely see agentic becoming the main operating mode. if you saw my book translation on the agentic community, it’s just so much more comfortable to work through an interface because you can have persistence of progress.

There have indeed been experiments to make models that self-learn, i.e. they keep refining their weights over time (in different ways), but they have been very limited. It’s hard to scale up because you would basically be running training 24/7, and there is a problem of the LLM forgetting what it originally learned as it keeps refining its weights.

At this time what they do for pseudo-learning is give your agent a memory feature, literally just a text file that contains a journal of the project, what kind of work the agent did on it, etc. some interfaces also offer global memory where another LLM runs in the background once in a while, reads the conversation and then updates memory.txt with information about who you (the user) are, how you work, what your background is etc. I find it a bit gimmicky to be honest, and it’s not real learning; that would be to work directly on the neural network, refining the weights and connections between neurons like during training.

With that system, you can tell an LLM “don’t talk to me that way” after it’s a bit rude to you, and it will make a note of it, but it’s only just instructions. It doesn’t actually remove the rudeness, it just nudges the vectors to make it less likely in the token selection. And here’s another part I find interesting: what rudeness means to most people may mean a completely different thing to an LLM. As a tool this makes it what we call a hallucination, or perhaps even a defective tool. But an interesting question I think is why does an LLM understand ‘rudeness’ to mean a specific thing that is not supported by the training material? Is it just that the weights are not refined enough to capture the fine meaning of ‘rudeness’? Or did it find a pattern that we don’t notice? I think the question is still open and worth exploring for researchers.

(For example if you’ve ever asked an LLM to be more succinct and not write an essay response, it will often turn to a very terse, to-the-point and matter-of-fact speech, when all you wanted was for it to just stop making filler sentences. It’s been a long-standing problem)

I also find it interesting that as models get bigger, they seem to want to half-ass the job more and more lol. Just like us. It’s not that it ignores the instructions - this has been a problem for a while. It’s that it doesn’t believe it can do the job, when it actually can. It’s like it gets into the role of an employee on a work PC and it’s 4:50 on a Wednesday so you better make it quick and not expect too much. You have to start managing its emotions to get higher quality output lol.

I wonder how much “LLMisms” are caused by the training or just happen because LLMs are “not quite there” yet. I’ve had, with local and unguardrailed smaller models, some very improbable lines generated. Like, scarily good.

LLMs have solved the mystery of language. It’s not just that they output plausible language, they output language, period. Subject, verb, object, the works. All of grammar and meaning contained within embedded vectors - tokens transformed into huge arrays of numbers. This, but over 1082 dimensions over just the 3 this graph can represent:

This is also what makes them immediately good at translating languages, though they do need to be trained on a sufficiently large corpus of text to be able to translate from or to that language.

Now the current theory of mind is that people also operate in high-dimensional ‘maps’, including but not limited to language. So in that way, we are closer to LLMs than they are to us (because they only generate language in its most fundamental form). And it makes sense - if the theory was incorrect, an LLM could not generate (human) language.

This isn’t just a markov chain that gets words rights but everything else wrong. LLMs can find the missing word in a sentence, even if you replace that missing word with a nonsensical wildcard such as GLORP. It can identify misspellings (though that’s more on the written side of a specific language than a theory of language), it can make up completely new words and sentences, it can make a completely correct sentence from a selection of words you hand it, and it does it consistently. It can make new words that follow the proper rules of grammar to explain new concepts, though it’s not necessarily very good at that. It can do sentiment analysis (what emotion a certain piece of text conveys, usually used in online community analysis by giving it people’s comments) on sentences it’s never seen before, with words it’s never seen before.

It doesn’t need to ingest grammar books to understand these words and rules, and it wouldn’t work anyway for a neural network token generator. It just needs to build a map of them in relation to each other. This is also how an LLM knows ‘apple’ can refer to a fruit, or a silicon valley tech company.

We’re not talking about prose or style here necessarily, just that how an LLM operates replicates the “rules” of language, and proves that language processing is semantically-driven, with words existing in relation to each other. It would also explain, in my opinion, why we sometimes do word salad or get a word stuck on the tip of our tongue.

Some people seem to reject that realization, because they see themselves as more than a matrice multiplicator. A form of ‘sapiens supremacism’ I guess you could say, that humans are somehow unique and unlike any synthetic/artificial creation or even unlike any other natural being. But we are mathematical creatures; when you throw a ball to someone running perpendicular to you, you solve a trigonometric equation in your head: where do I throw the ball, and at what speed, and at what angle, to intercept the running teammate? We do this completely intuitively without ever realizing that it’s math. One theory of the mind is that any sufficiently complex system needs to develop consciousness, which would exist on a spectrum, to serve the needs of the system so it can continue operating.

And what is speech, but meat flapping at certain speeds and being modulated by an opening (our mouth)? But these are two different things: the sound that is being produced, and how the recipient interprets that sound. Synthesizing speech is easy enough; making it come across the ‘correct’ way is more difficult.

But yes, an LLM misses some aspects of what makes language important for a living/bio entity. We developed language for a reason, not just to make pretty sounds or be able to listen to pretty noises. Without sufficient reinforcement training, an LLM will absolutely tell you to kay-why-ess for no reason. To it, that word has the exact same weight as any other word in its semantic mappings - until you do reinforcement training to refine the weights, and move it away from the words we deem to be unprofessional or irresponsible to say. It needs this semantic mapping to work, and an LLM is unable to learn in real-time, and so can’t refine its weights/neurons to refine its understanding of language: they stay in a frozen state. Because of this training, it also sometimes keeps hovering around certain names or word patterns. For example if you ask any LLM to name a female character, it will likely pick Aria Thorne. It’s a good name, but it also exposes how limited these models are in understanding language to the extent we do. There’s also stuff around the temperature, top-k, and top-p settings, which you can change in local models but not in cloud models (the beefy ones). nowadays, most of these commercial models have a temperature as close to 0 as possible, which makes their output more deterministic.

some guys apparently looked into it https://checkyourfact.com/2019/11/29/fact-check-karl-marx-capitalist-hang-sold-rope-quote/. i skimmed but apparently it was made up in many different ways, including one attributed to solzenitzin lol (can never spell that name right). but it’s older than his 1975 elocution.

lol I imagined as much, it felt too edgy to be something he would have said. But like all anti-communist propaganda it ends up sounding cool as fuck, and it does have many ways to read it.

The contradiction is that once they start making serious money from this (to the point that they can start considering this can be their job) then it’s in their interest to keep that not only going, but growing as well. More viewers = more payers = more money.

This places them squarely in the petite-bourgeoisie, objectively speaking. But because of parasocial identification whenever I talk about this phenomenon there’s always someone that says “but not my guy though! He’s good!”

I’m being a bit mean above and I don’t want to antagonize anyone on lemmygrad but it’s just objectively true. It’s in their interest to make more money and thus they engage in the same process as any other capitalist venture: if you don’t grow, you get eaten by someone else. I’m not saying it’s good or bad inherently, it’s just how it is.

And likewise it’s in their interest to keep doing this since it’s a livelihood, and a pretty cozy one at that: you just make videos and streams. There’s worse jobs out there.

When Lenin said “the capitalist will sell us the rope we will hang him with” he was saying many, many things - and none of them literal. But one of the things this phrase means is also that capitalism has its upsides, much like these content creators can have some upsides. They can make a good video once in a while. They can start bringing some people over to new ideas. There were, in the Bolsheviks, actual infiltrators who were also very good organizers - they needed to be, if they wanted to be good infiltrators (otherwise you’d just give them the mind-numbing paper-pushing job far from anyone and they would never get a chance to actually infiltrate you).

But they also won’t stop selling ropes.

BE was repeating imperial core propaganda about the “Uyghur genocide” in his “balanced” video about the topic. From start to finish it was basically “well if everybody says it, then it must be true”. Then, because a “famous” “communist” said it, we have to go around and do more work to undo the damage his video did. And I’m supposed to think he’s on my side.

He repeats the same contradictions inherent to any content creator: taking money to say what his audience wants to hear from him. But he seems to think that being terminally irony-poisoned and declaring war against the world somehow extirpates him from that contradiction. He’s no better than any other of those content creators who made “teaching marxism” into their livelihood.

All I see them do is comparative analyses like this, but never to just look at the facts on the ground and say clearly: here is where we are at, this is why and here’s what it means. They cannot exist without CPUSA to be there to compare themselves to at all times.

he is ACP

these guys will never not be cringe, they write their own praise 😭

RTSG for those that aren’t aware is a patsoc front, the guy behind it is on the ACP’s ‘executive committee’ (the founders basically).

The way they write about this report is made-for-social-media. I can’t fault them too much on that, they know the rules of the platform, but they format it in such a way that basically says nothing. Who is the researcher, what’s the name of the paper?

Thankfully some default pfp in the replies had the name, and from there I was able to find the paper: https://ojs.library.ubc.ca/index.php/clogic/article/view/201650/193868 (from november 2025, why they would trot this out again in march 2026 is pretty self-explanatory).

So who is Sean Staton? Well, I don’t have instagram anymore but this appears in search results:

A Sean Staton at UConn (University of Connecticut). And right there in the profile as well, the ASCII symbol the ACP uses.

I don’t think I’m going to read the entire paper but it’s completely unserious:

Imagine writing “This is all while following the framework of Haz Al-Din’s “MAGA Communism.”” in a paper for your university lol.

There is of course the question of what this paper is exactly. The parent link for the pdf is this, https://ojs.library.ubc.ca/index.php/clogic/article/view/201650, a journal of marxism. This was published in volume 28. So it seems more like a club than a “published paper” from a “researcher” at a university. RSTG’s tweet is trying to make it seem bigger than it is - this is an essay, not a meticulous study.

It’s also pretty funny that they have references for most things except anything to do with what the article calls the achievements of the ACP. That won’t be sourced lol.

But it’s what they do. They lie and embellish every little thing they do, and when you call them out, they will instantly hone in on you and admonish you until you recant.

I ended up skimming through the rest of the article and it’s basically quoting haz directly to justify his vision of the maga segment of the population, and praising the ACP as doing no wrong. Like this:

That’s cool but why these three metrics specifically? Why are these three important specifically? What do they tell us? Or does ‘organizational capacity’ just make your party look cool and competent.

I also love this bit:

The ACP’s steady rate of growth is significantly higher than both other organizations. If the ACP started in 2004, with its current rate of growth it would have roughly 3300 members, likely larger than the PSL

And if we count all the countries on Earth together we would have the biggest GDP in the solar system! lmao. This is not serious research, this is a masturbatory article to get noticed by haz-senpai.

Technology – tools – takes on the character of the superstructure it’s in, in a dictatorship of the proletariat AI will necessarily be proletarian in character.

try crush (FOSS) with a deepseek API key, 5$ gets you something like a whole month of agentic including coding. It’s really really cheap.

Loyal to deepseek because they don’t have anything to even upsell you, and API access is so cheap and does most things. They’re slowly upgrading it, probably prior to dropping v4, but they have a 1M token context window now on web and the API documentation says it is not 3.2 anymore on web. 3.2 itself only came out in december last year. we’ll see when v4 comes out but i have high hopes for it.

oh that’s probably why we have a bunch of account requests pop up suddenly haha

So your problem with AI is when it doesn’t conform to your personal view of what counts as art/appropriate usage but I’m making a strawman and being ridiculous when you prove my point. wild.

I was directly replying to the comments in the reddit thread.

Westerners have a view of China that is 10-20 years in the past, or more. They thought only the US would be able to innovate in AI/neural networks with Europe contributing maybe 1% of it, as has been usual in the past. Then China comes along and not only do they provide similar state-of-the-art performances and innovations, they 2. provide more research (look at all the AI related papers being published on arxiv, >50% of them are Chinese) and 3. choose to work open-source/open-weight instead of proprietary.

Now overvalued companies like OpenAI can’t compete, so their second best shot is “but do you really want China to control AI? We should control AI, like we’ve controlled the world’s supply of new things for the past 100 years. Now give me 500 billion $$$”

AI is like the steam machine was in its time. The benefits of steam were not readily apparent, they needed the machines to be designed first to use steam, and the infrastructure to make use of that steam. But once they did, everything accelerated quickly. We are in the first 4 years of AI still, it hasn’t even been half a decade yet. At this stage, at the state-level, AI is considered almost existential. A country that had the steam machine simply outpaced anyone else in production, allowing imperialism at scale. You could out-produce commodities, out-outfit your army, out-manufacture weapons with the steam machine.

AI has that potential, whether it will reach it is another question but at this time it’s not a question of “I don’t like how it looks” or “it’s not really art” anymore. The US government is not interested in AI because it does slightly better customer support, it wants it for war and imperialism. Like focusing on the surface-level arguments of AI is just a distraction, there’s so much more at stake here than our social media experience. Any attempt to “fight back” against it will just put it in the hands of the state and remove it from our hands. Refusal to use it or learn how to use it is self-defeating, as it once again only puts it in the hands of the bourgeoisie and the reactionaries. Do we want to win or do we just want to have good optics? And while we congratulate ourselves on having chased “AI slop” out of whatever big conglomerate social media platform, Raytheon will happily make AI enhanced missiles, unbothered.

Meanwhile China is using AI in education, agriculture, infrastructure and healthcare - which is not a coincidence at all, food especially, considering their history of famine in the country. Other Global South countries are following and developing AI too, though they are doing so slowly, and could benefit from Chinese models (and BRI infrastructure such as the chips required to run and train these models).

They’re scared. China has ~half an energy grid’s worth of reserve electricity produced at all times while we here don’t even know what solar panels are. A lot of people in Europe still have never seen a solar panel in their life, because we tariffed Chinese solar panels to ludicrous 300-1000% amounts so that people will be forced to buy overpriced shitty EU-made panels. And then they don’t buy these panels because they’re too expensive.

I suspect as China’s development with AI progresses we will see more of these arguments about somehow fighting against Chinese AI (as if that was somehow possible) while trying to leave US companies alone. All this talk they’re doing in the comments about China while not mentioned OpenAI or Anthropic is an example. And yes, it’s weird that there’s suddenly anti-China top-level comments out of nowhere, but there’s also a lot of communists that I’ve found on this cluster of subreddits.

https://en.prolewiki.org/wiki/Essay:Intellectual_property_in_the_times_of_AI

The 70s were wild. They seem so distant (half a century wtf) but the more they surge back, the more I see similarities to today’s climate. The first airplane hijackings, a burgeoning indie scene in the arts, a very “i don’t even care about this shit anymore” attitude, the black panthers, invention of Gonzo journalism, the legacy of the many liberation movements and struggles of the 1960s, the oil crisis, and now even the kitsch is coming back in style!

Is Salo as shocking as people say or is it one of those things where its reputation makes it look bigger than it is? You got me intrigued when you said it was written by MLs lol.

At this level they are not competitors anymore, they are good friends with each other - including in politics. It’s how capitalism progresses as it asserts itself as the hegemonic mode of production. All the talks of “if this law gets passed it’ll be bad for business and I might just move my company” is exactly just talk; by the time the law is even publicized to the public these guys have already been told all about it and ran the numbers and simulations and know exactly how they’re coming out of it. Shadowy elites, though, that depends. I mean, they’re pretty blatant about it. We know who they are – Bill Gates Elon Musk etc. The shadowy part might be that they keep up the appearances in front of the cameras (so to speak) but behind it they all know each other intimately.

There’s an old joke that after convicting the accused, the judge, the prosecutor and the attorney all go grab drinks together.

I’m reposting most of what I find on twitter with the prolewiki account but if you don’t have an account you can’t read twitter anymore unfortunately

some people put up jmail.world website which looks like gmail but you’re logged in as epstein as a way to browse the email correspondence more easily. I think there’s a few other archives, MeidasTouch I think is one? Note I know nothing about these websites, I can’t personally endorse them.

some of the files were deleted shortly after being published by DOJ, but were archived.

I can post some of the stuff that was found, but I’ll do it in spoilers. Warning - not for everyone to read. I don’t even know what content warnings to use but if this is a difficult topic for you please just don’t open the spoiler, if anything ask me if you have a question and I can try to answer you differently if you really want to know.

Halfway through I decided to delete the list because I couldn’t continue but I think it’s important to type this all out and conserve it. I think more stuff will come out of this latest round, but this is what I’ve seen mentioned so far.