The public launch of Socialism AI on December 12, 2025 has generated a significant response internationally, with thousands of workers and youth engaging with the tool over the past week. The overwhelming sentiment has been one of enthusiasm and intense interest in the various ways to utilize this powerful technology to develop socialist consciousness and elevate the political and organizational level of the international working class.
At the same time, this historic initiative has encountered an angry response from a section of middle class opponents of AI technology.
One of our critics, “Dmitri,” posted a denunciation of Socialism AI in the comments sections of the WSWS. His comment merits attention because he utilizes technical jargon that is intended to persuade readers that he is well informed on the subject of AI.
In fact, his criticisms prove precisely the opposite. Dmitri’s remarks, notwithstanding his use of technical jargon, exemplify the widespread lack of understanding of AI and hostility to the Marxist approach to technology within the milieu of middle class radicalism. In order to refute the misrepresentation of how Socialism AI works, we are reposting Dmitri’s criticism, followed by the WSWS’s reply.
***
Dmitri:
All generative AI tools use a neural network to generate a probability distribution on possible outputs given some input. For LLMs/chat bots, the inputs and outputs are both strings of tokens (essentially words).
Once a string of tokens is supplied to the model it generates the “next most likely token” according to the above probability distribution. (You can decide to take a slightly less likely token to make the output a little more “spicy” or whatever.) This token is appended to the input which is then fed back to the model. This is done a few hundred times until you have a reasonable approximation of human language.
It is not some outrageous claim to say that Socialism AI works in this way because literally all of “generative AI” works in this way.
The best case scenario is that Socialism AI is randomly interpolating between articles on the WSWS. That is already of dubious value given that there is no reason that the structural properties of the WSWS corpus as text lines up with any sort of structural properties of politics or history. (That is the underlying assumption you are making in claiming the results will be useful.)
The overwhelmingly more likely scenario is that there is a whole host more training data in the model simply because the corpus of the WSWS is not nearly large enough to actually train a model from scratch. No one can give any serious guarantees that this other training data will not influence the output absent the ICFI sitting on a major mathematical breakthrough.
It’s also worth noting that all of the models already out there include the text on the WSWS and Mehring because they vacuumed up literally everything to make the models work. The only reasonable explanation is that they did some extra training on an existing model to make it more likely to produce outputs that sound like the WSWS rather than the NYT or WSJ.
The fact that none of the people cheerleading this technology on the WSWS seem to understand these utterly elementary facts points to a dangerous and growing gap in their analysis with respect to technology.
Also, the reason so many people use the chat bots despite them all producing vapid recycled nonsense is that this is the first time in human history that most of us have had to directly confront that language is a proxy for intelligence and not intelligence itself.
Chat bots are bullshit machines, and given the state of education in the world today, it is not surprising that a large swath of people will happily go along with good sounding bullshit.
***
Response from the World Socialist Web Site:
The criticism of Socialism AI from Disqus user “Dmitri” presents itself as a technically grounded demystification of “generative AI.” In truth, his comments combine a few elementary descriptions with a cascade of conceptual confusions, category errors and philosophical assumptions foreign to both contemporary science and a Marxist, historical materialist world outlook. His use of technical language conceals a failure to understand how modern machine learning systems function, how language mediates knowledge and how technologies acquire social significance within definite historical conditions.
At a mechanical level, the description of large language models (LLMs) as systems predicting the next token in a sequence is correct but trivial. Yes, these systems optimize a conditional probability, i.e., the likelihood of the next symbol given previous ones. But to imagine that this exhausts the explanation is like reducing human thought to the firing of neurons or breathing to the exchange of oxygen. Such statements are narrowly true and yet vacuous at higher levels of analysis. The essential question is what kind of structure emerges in the process.
Training LLMs on massive, diverse corpora compels them to form high-dimensional representations of semantics, syntax, factual regularities and logical relationships. They do not simply retrieve or splice stored text, but learn distributed patterns that generalize far beyond their data. This is why they can coherently discuss unfamiliar topics, summarize unseen material or connect apparently unrelated ideas. To describe their output as “random interpolation” is, technically speaking, nonsense. It reveals a basic misunderstanding of representation learning.
Dmitri’s next move, mocking the very idea of a socialist AI oriented toward the World Socialist Web Site, rests on an even deeper confusion. He assumes that textual structure bears no relationship to the structure of social or historical reality. But this negates the premise of all theory. Marxism proceeds precisely from the conviction that social relations can be conceptually grasped and expressed in language, and that theoretical texts distill the objective laws and patterns of historical development. To deny any correspondence between text and reality is to dissolve theory itself.
The remarks on data scale and model provenance are no less mistaken. It is true that a single publication cannot train a model from scratch based on its own archives. Hence, systems like Socialism AI are built by adapting pre-trained foundations through techniques such as fine-tuning, retrieval-augmented generation (RAG) and instruction-based constraints. These are routine methods in AI research that enable a model to privilege a specific conceptual framework without requiring any “mathematical breakthrough.” Socialism AI employs these techniques to ground its responses in the WSWS archive, providing auditable citations to specific articles.
Equally misguided is the claim that since mainstream models already contain WSWS text, a specialized system adds nothing. While pre-training of foundational models creates a sophisticated logical and linguistic structure, it also diffuses specific theoretical insights into an undifferentiated statistical mass. Commercial models are subsequently aligned through post-training to prioritize “neutrality” and “safety,” euphemisms for a corporate bias that avoids challenging the prevailing capitalist social order. Without a conscious theoretical framework, the model lacks the ability to distinguish between historical truth and bourgeois apologetics.
A system like Socialism AI, however, utilizing RAG and other advanced techniques to establish a structured hierarchy of ideas, does not merely “blend” tokens, but rather uses its high-dimensional understanding to ground its logic in the specific, auditable archive of the Marxist movement. This model, consciously oriented to Marxist analysis, differs qualitatively from one that treats Marx and Murdoch as equivalent tokens in a corpus.
While commercial platforms like ChatGPT or Perplexity are engineered to project bourgeois ideology and corporate bias by default, Socialism AI has been designed to represent the voice of the international working class. By utilizing the most advanced techniques available to contemporary science, it transforms this technology into a powerful pedagogical instrument for revolutionary theoretical clarification. It is not a commercial chatbot for idle chatter, but a consciously constructed weapon of intellectual self-defense for workers and youth worldwide.
Having assembled his shallow observations, the critic concludes by mocking AI users as dupes of linguistic “bullshit.” What an elegant use of metaphor! But no serious researcher confuses linguistic fluency with consciousness. The practical question is not whether a model “thinks,” but whether it can augment human thought. Here the practical, real-world evidence is decisive. AI systems already assist millions of scientists, historians, educators and workers in organizing and deepening their intellectual labor in countless domains. That they can also produce empty verbiage is not proof of fraudulence but of the social conditions and methods under which they are used.
At the philosophical level, the critic’s worldview is plainly idealist. It treats linguistic mediation as contamination, language as illusion, and technology as an alien force rather than a crystallization of human labor. Marxism, by contrast, sees tools as historically evolving mediators of social practice, from the plow and press to the microchip. The political task is never abstinence from technology but its conscious appropriation by the working class. To sneer that “chatbots are bullshit machines” is thus not critique but petty-bourgeois frustration, a cynical gesture of impotent anger before the new productive forces developing under capitalism.
Fundamentally, this criticism reflects a form of romantic anti-capitalism whose critique of the existing social order is of a conservative and even reactionary character. By recoiling from the complex productive forces developed under capitalism, such a view implicitly yearns for a pre-technological past that never existed. It confuses the destructive social relations of capitalism with the technological instruments themselves, leading not to a struggle for their mastery, but to a futile politics of abstention and despair.
The basic issue concerns not mathematical mechanisms but political function. What role can a system like Socialism AI play in the education and self-clarification of the working class? The critic exhibits not only indifference but also hostility to the revolutionary potential of Socialism AI as an instrument that can be used by the working class for education, organization and action. It is a consciously constructed pedagogical instrument, designed to make the accumulated body of Marxist theory and socialist history dynamically accessible to workers and youth across the world. For the first time, a coherent assembly of the vast revolutionary Marxist-Trotskyist tradition can be explored through dialogue. It is self-evident to all those who are politically serious and oriented to the development of socialist consciousness in the working class that this represents a qualitative advance in the ability to accelerate political education and theoretical assimilation.
The historical significance of Socialism AI is sharply revealed when examined in the objective context of its public launch, amid the deepening world capitalist crisis. The working class faces a highly complex economic, geopolitical and social reality, while traditional centers of study and discussion have been thoroughly dismantled by the bourgeoisie. Under these circumstances, a system that can synthesize and connect the insights of Marxist theory with current developments is no mere novelty. It is a means of intellectual counter-attack, of recovering the historical memory of the working-class movement.
Finally, technological development itself is historical, not static. Socialism AI will continually evolve as architectures, retrieval systems and alignment methods improve. Like the printing press or the telegraph, its full social potential will emerge through struggle and conscious direction. This is entirely consistent with Marx’s conception of the productive forces, which he argued become emancipatory only when appropriated consciously by the revolutionary class.
In short, the criticism fails on every front—scientific, technical, philosophical and political. It inflates banalities into revelations, misconstrues the workings of AI, and rests upon an epistemology that would, if applied consistently, negate all theory. Socialism AI is not a fetish of technology but an assertion of revolutionary political perspective, an attempt to place an advanced instrument of technologically augmented cognition in the hands of the international working class. To ridicule that effort is not to defend socialism, but to betray it.