Artificial Intelligence or Artificial Illusion?
July 24, 2026
Long story short, and as we flagged in the last Market Update, several members of our team and I spent the better part of a year taking the AI story apart piece by piece, peeling back one layer after another until we could finally separate the genuine technological progress from what turned out to be one of the most sophisticated marketing campaigns Wall Street has staged in decades, and the conclusion we landed on was almost embarrassingly simple, because beneath the breathless headlines, the trillion-dollar valuations, and the endless promises of a new industrial revolution sits an extraordinary quantity of Horse ST**, wrapped snugly around a technology that is genuinely impressive yet routinely dressed up as something it plainly is not. That distinction matters far more than it might first appear, since every great technological revolution eventually arrives at the very same crossroads where one path is driven by engineering and the other by marketing, and history keeps demonstrating, almost monotonously, that marketing very nearly always outruns engineering, whether we are talking about railroads, radio, aviation, the internet, blockchain, electric vehicles, or now artificial intelligence, because although the underlying technology changes from era to era the human psychology surrounding it never does, so that investors stop valuing what actually exists and start pricing whatever they imagine might someday exist, and somewhere in the gap between those two worlds reality quietly slips beneath expectation.
Large Language Models Live in the Gap
Large Language Models sit squarely inside precisely that gap, and the public faces of the industry, whether that happens to be Sam Altman, Dario Amodei, Elon Musk, Mark Zuckerberg, or any of the countless AI evangelists working the circuit, have managed to convince both investors and the wider public that LLMs simply are artificial intelligence, which they are not, since they represent only one branch of AI and are far more accurately described as extraordinarily sophisticated neural networks capable of recognising patterns, predicting probabilities, and generating language that often looks remarkably intelligent, an achievement that genuinely should not be understated because it ranks among the most important engineering breakthroughs of the past several decades, and yet calling statistical prediction genuine intelligence is rather like calling a calculator a mathematician, given that the output may well be useful while the underlying process remains fundamentally different. The problem was never the technology itself but rather the mythology that has grown up around it, because today’s models do not understand the world in anything resembling the way human beings understand it, possessing neither intention, self-awareness, curiosity, nor genuine reasoning independent of the data on which they were trained, and instead simply predicting the next most probable token based on patterns absorbed from unimaginably large datasets, so that when those patterns happen to align with reality the results can be extraordinary while when they do not the very same systems confidently generate errors, fabricate sources, and produce conclusions that persuade precisely because they imitate confidence so effectively, which is not intelligence at all but statistical fluency operating at an astonishing scale.
The People AI Helps Most Are the Ones Least Threatened by It
There is a real irony buried in all of this, which is that the people who benefit most from AI tend to be exactly the ones least threatened by it, because if you are already highly competent in your profession these models become remarkable productivity multipliers that accelerate research, organise information, challenge your assumptions, and strip away countless hours of repetitive drudgery, so that used well they let experienced professionals spend far more of their time thinking and far less of it searching, whereas used badly they do nothing more than accelerate ignorance. To put it bluntly, if you are not already competent in your field and you lean on AI blindly, it is about as useful to you as a barn mouse, because without the experience to weigh its answers, catch it when it is wrong, or tell genuine insight apart from plausible nonsense you have essentially handed your judgement over to a probability engine that has no comprehension whatsoever of the consequences flowing from its own output, which is why AI does not eliminate the need for expertise but instead makes expertise even more valuable, since someone still has to recognise the difference between a convincing answer and a correct one.
The Great Economic Reversal Nobody Expected
That may well turn out to be one of the great economic reversals of the coming decade, because for years we have all been told that experience matters less than adaptability, that institutional knowledge is a relic, and that younger generations raised alongside technology would naturally displace the very people who built the industries they now stand to inherit, and yet reality is shaping up to be considerably messier than that tidy story, since experience is actually becoming more valuable precisely because AI amplifies existing competence rather than conjuring competence where none previously existed. The professional carrying thirty years of accumulated judgement suddenly acquires an assistant capable of processing information at extraordinary speed, while the inexperienced worker acquires the identical assistant yet lacks the framework needed to evaluate anything it recommends, so that although the machine remains exactly the same in both hands the outcomes diverge wildly, and that observation ought to feel familiar because it reflects a principle we have returned to repeatedly throughout these pages, namely that technology rarely replaces judgement but instead amplifies whatever judgement already happens to be present. Hand the same AI system to an experienced surgeon, an accomplished engineer, a seasoned investor, and a complete novice, and you will not get four identical results but four entirely different ones, because the decisive variable was never the model at all but the quality of the mind directing it.
Advanced Imitation Dressed Up as Intelligence
All of which brings us to perhaps the biggest irony of the lot, since AI is supposed to stand for Artificial Intelligence and yet, judging by where the technology actually sits today, a more honest label might be Advanced Imitation, or possibly Artificially Impressive, because what these systems genuinely excel at is not understanding but imitation, and they imitate language so convincingly that people routinely mistake fluency for comprehension, confidence for reasoning, and probability for intelligence. The engineering behind these systems deserves enormous respect, but the mythology draped over them deserves a great deal more scepticism, and markets have never been especially good at telling the two apart, because every major speculative cycle eventually reaches the point where the story becomes worth more than the underlying asset, at which stage investors stop asking what the technology can actually do and start asking what they hope it might one day become, which is almost always the precise moment when narrative overtakes evidence and marketing quietly begins generating fatter returns than engineering ever could.
Spending Has Soared While Returns Have Stalled
Artificial intelligence is unlikely to escape that cycle for the simple reason that no revolutionary technology ever has, so the long-term winners will almost certainly emerge, real fortunes will undoubtedly be made, and genuine breakthroughs will keep arriving, and yet history strongly suggests that most of today’s promises will quietly evaporate long before those breakthroughs fully mature, which may be the real lesson here, because the first generation of LLMs genuinely earned the excitement it received as a meaningful leap forward while much of what has followed has proven far less revolutionary. Hundreds of billions of dollars have been funnelled into ever larger data centres, GPU clusters, and marketing campaigns, and yet the improvements have grown steadily more incremental even as the narrative keeps promising exponential progress, and China has already shown that smarter engineering and greater efficiency can close much of the gap without simply hurling unlimited hardware at the problem, which exposes an uncomfortable truth a good many investors would clearly prefer to ignore. The technology itself has not stalled, but the returns on spending most certainly have, and one could reasonably argue they have been dismal at best, while the market carries on pricing AI as though every additional dollar invested will inevitably conjure another revolutionary breakthrough, which is not engineering but pure narrative, since the first wave was genuine innovation whereas much of what has come after looks far more like capital expenditure desperately searching for a justification, and somewhere between the trillion-dollar valuations, the endless GPU orders, and the assurances that true intelligence is always just one more model away, the AI story has grown increasingly hard to distinguish from Horse S**T.
The Insightful Journey to Profound Understanding
