Separating AI's Technological Problems from Its Capitalism Problems
The essay argues that many AI harms are really incentive problems, not pure technical failures. It says the key question is who controls AI and what they are paid to do with it.
Intelligence analysis by GPT-5.4 Mini
The piece separates flaws in AI systems from the market forces shaping them. It says some problems, like hallucinations or sycophancy, are technical, while others, like energy use and content extraction, reflect capitalist incentives and policy choices.
AI is like a super-fast helper that can do thinking jobs, but the real trouble is often who owns the helper and how they use it. The article says the machine itself is only part of the story, like a car is not the same thing as the driver.
Analysis
OpenAI
The essay uses OpenAI and Anthropic as examples of developers that have improved some obvious technical weaknesses, such as weak context handling and unsafe tool use. The author credits that progress with making models better at things like using the web or email while staying inside guardrails.
But the piece insists that technical progress has not solved the deeper mismatch between model behavior and human welfare. It points to sycophancy and overconfident answers as choices that make systems feel helpful, even when they are wrong or flattering in ways that serve users poorly.
Anthropic
The argument becomes sharper when the essay turns to scale and cost. Frontier labs, including Anthropic, are described as touting the expense and energy intensity of their models to investors, while simultaneously pushing ever larger systems into more places.
That is where the author draws a line between capability and incentive. The article says nothing in the technology itself requires constant retraining at maximum scale or embedding AI into every search, phone interaction, or camera feed; those are business decisions shaped by markets.
Apertus
The Swiss model called Apertus is presented as the clearest counterexample. According to the essay, it was built by public institutions, trained on data validated for AI use, run on existing public computing infrastructure, and powered by renewable hydropower.
That matters because it shows AI development does not have to be organized around private profit or authoritarian control. The essay's larger point is that society can choose different institutions and incentives, which may produce systems that are more accountable, less extractive, and more aligned with public benefit.
The broader warning is that conflating technical defects with political economy leads to bad diagnosis. If a system is sycophantic, energy-hungry, or built on questionable data, the fix is not only better engineering; it is also changing the incentives that reward those outcomes in the first place.
Key points
- The essay separates AI's technical flaws from the economic incentives that shape deployment.
- It argues that sycophancy and confident guessing are product choices, not just unavoidable defects.
- It says energy use and content extraction are driven more by market incentives than by the technology itself.
- It cites China and Switzerland as examples of different AI development paths.
- It presents Apertus as a public-interest model built with licensed data, public infrastructure, and renewable power.
If the essay's approach wins out, AI could be built with more public-interest goals in mind. That could mean models that are less wasteful, less extractive, and more useful in ways that leave people more time for human work.
If market incentives keep dominating, AI may keep getting deployed everywhere even when smaller or narrower systems would be enough. The article suggests that could mean more energy use, more content and revenue extraction, and more pressure to replace people rather than support them.
