One striking feature of contemporary “artificial intelligence”, whatever that might mean today, is that it is somehow nothing and everything at the same time.¹ On the one hand, it demonstrates impressive feats of aggregation² and compression;³ on the other, it fails spectacularly at tasks of both logic (expected, given the technical structure of large language models) and recall (underexpected, because wasn’t this one of the defining “advances” made by AI?)⁴ The result, then, is a temporal distention, where generative AI in 2025 is kinda already here but also always only really arriving in the future — either via improvements or via the holy grail of AGI. The latter part of this temporality can be understood as a scalar movement through the speculative, where the next-word prediction (of an LLM) and the next-moment prediction (of financial investments) find themselves entangled in the look to the future: what may happen? What may be wanted?⁵ An important part, however, of this speculative dimension is one that is embodied in the user: the subject par excellence of the service economy.⁶ If all of us are users — and if this has somehow superseded our being citizens or workers (big if) — then perhaps it is worth asking what kind of user this particular user (of generative AI) is.

I want to argue here that this user is a beta: a tester, first and foremost, even before they get to be a consumer. Further, this “being beta” marks a specific, testy relationship between citizens of the global North and the matrices of extraction that they are caught in today.⁷

It might be evident to some readers how we are all testers today (some of us more than others, of course). As tech companies fail to generate the profits they desire from infrastructurally expensive chatbots, cheap VC-funded chats with LLMs end up underscoring how generative AI is truly a solution in search of a problem.⁸ Each interaction with an LLM, then — be it awe-inducing or underwhelming or somewhere in between — is an attempt to automate (part of) some other labour pipeline. In doing so, it also becomes an attempt to gather more data on said pipeline, while the whole world throws the kitchen sink at a piece of technology that has a very specific modality — one that masquerades as an all-purpose suggestion of a possibly useful tool.⁹ It is true that there is often some alpha testing that takes place inside company/lab offices before a new model is released; but by and large, it is clear that the mantle of beta testing now falls squarely upon all of us.¹⁰ This has a long recent history: companies once outsourced private R&D and the development of intellectual property to the much cheaper, subsidised public education system;¹¹ they outsourced the labour of writing code to the global South;¹² and they crowdsourced insights and data collection to all of us, and to our sociality by extension, making us complicit in the very production of the machines that we use.¹³ Today, they outsource not just the personal, cultural, and social implications of their product (the so-called alignment), but also the very product-ness of their product (what it is for, what it does, what it cannot do) to millions of beta testers today.¹⁴ If we cannot rely upon the given product to do what we want it to do (or hope for it to do, or delude ourselves into thinking it can do), then it contributes to the general precariousness of our existence under late capitalism, while also reminding us how some technologies were never tools — and perhaps could never be(come) tools.¹⁵

The subjective move from being a user to being a beta tester/user does not just tell a straightforward tale of expanding work-as-precarity, but also signifies the present-day social relationships between technology and extraction writ large.¹⁶ At the heart of this move is how extraction is enabled across distributions of time and space. Two simple terms-as-models can help elucidate what is not being straightforwardly discussed here. The first is extraction (of resources, as surplus value or via explicit violence) from one part of the world (say South) to another (say North); the second is accumulation (of resources, such as by means of enclosure), and the entrenchment of a given state of materialities. What is at stake here is instead the act of imagination of an operation — what even is going on when a user meets the so-called used product? — that is being extracted and accumulated.

To refract via some conventional Marxist frameworks, the user is being asked to consider possible reifications, and to hand over the blueprints for the same to the capitalist.¹⁷ The user sits down at the screen and thinks step-by-step with the machine, a common prompt-interface modality, and throws possible use-cases at the model — tracking the efficacy of the fledgling cyber-homunculus,¹⁸ guiding and coaching it into usability, into getting better at tasks that will one day, in the future, be automated.¹⁹ Perhaps the real agentic AI was the users we agent-ified all along.

Allow me to ruminate on the beta-ness of it all. On the one hand, as a compulsory beta tester, every user makes deals with the incompleteness and the ongoing-ness of the state of affairs around us. If something is broken or wrong at the moment, it is because it will be fixed in the future (by more of the same).²⁰ Even if a promise was flawed, in the moment of interaction, maybe it’s because we — the betas²¹ — didn’t try hard enough?²²

On the other hand, a beta tester/user also emerges as a beta, the perpetually embarrassed loser in the folk pseudo-socio-psychology that is the wolf-pack nonsense. A beta, in this sense, is someone not atop the pecking order; a static mirage observed by some biologists, who themselves now realise how familial solidarities are a more complicated affair in animals and humans alike. But our beta user never got that memo from ChatGPT. He — and now it is a he — still wants to be an alpha; and a sigma grindset, leading him to higher productivity through automation and assistance, is his way to get there. But in this desire, too, he remains a beta, with LLMs undercutting his productivity, while at the same time, increasing the speculative productivity of a future market thoroughly infused with GAI or AGI. Like all stupid pop-psych garbage, the beta as a marker remains typologised unfairly, but in a way that structurally prohibits the move out of the stereotype. Always already a beta, tells the structure to the beta user. And to think: this beta user is not even the real worker, who is elsewhere — in the global South, inside the factories of materiality.²³

And yet, there is revenge in the offing. Alongside the secret sauce of the operationalisation of intellectual processes (or not-so-secret; in most cases, it is just tacit knowledge being articulated symbolically), the beta user ends up polluting the very well that holds his extracted insights.²⁴ As is being demonstrated by the recent flagging results from bigger and newer LLMs — and the monster of scale could certainly never have been slain so easily — the very fact of knowledge extraction comes fully equipped with its own dialectical movement: the extraction of ignorance.²⁵ In this global monkeys-on-a-typewriter experiment, OpenAI (and Google, and Meta, and so on) expect an eventual convergence between experts and expert tasks; in other words, the companies assume and hope that if enough experts train our systems for long enough, our systems will one day exhibit the same expertise.²⁶ But simply because most of what we do online, or on computational media writ large, is stuff that we have little clue about, the beta users end up conveying even more of wrong-ness and doesn’t-work-that-way-ness than of what is actually right, or of how to do or evaluate something. In this regard, the “we do not use student data” move by AI companies should not be read solely as a legal arse-covering (even though it is one: several legal frameworks have strict codes about what information can and cannot be shared outside a prescribed educational environment). It is concurrently also an attempt to channel away the worst pollutants of this future well of (reification) wisdom: the students who are clearly still learning, and often simultaneously trying not to learn, as all learning is, by definition. By clearly marking such interactions separately, corporations hope to rejuvenate the dying model-cycle, which was already showing clear signs of decay — either by disease (of existence), or by knowledge pollution, or through the antinomies of synthetic data.²⁷ If not the final, then perhaps the penultimate laugh — a maniacal laugh — is the beta user’s; a moment of latent sigma-fication.

The true sigma move, in this (mildly) new set of social relations, then, I argue, is to be as stupid as is humanly possible. I promise to do my bit. Will you do yours?