AI vs AI: the race with no finish line, with Eleftherios Jerry Floros

Why this conversation matters now: #

Most of the AI conversation in fintech and beyond is running in one of two registers. The optimistic register — abundance, productivity gains, cures for disease. The apocalyptic register — job displacement, existential risk, doom. In this episode of the Fintech Garden Podcast, Eleftherios “Jerry” Floros — author, keynote speaker, and one of the more independent voices in the current AI debate — offers a third register that neither camp is running with. Jerry calls himself an AI realist. Not an evangelist. Not a doomer. Someone paying close attention to the economics, the physics, the politics, and the human cost — and drawing the conclusion that the current trajectory is unsustainable, but not for the reasons either side is arguing. The conversation with host Igor Tomych is direct, structured, and — refreshingly — grounded in numbers rather than narrative.

What AI actually is, and what it isn’t: #

Jerry opens with the honest observation that even the people building AI cannot agree on a definition of it. His working definition is deliberately narrow: AI is when a computer program can imitate human intelligence in specific ways. There is narrow AI, which does a single task like playing chess or protein folding. There is what the industry calls AGI, general AI, which would do many things. And there is the philosophical question of whether AI, trained on data scraped from the internet, can ever replicate the nuance, critical thinking, judgment, and common sense that human intelligence carries. Jerry’s answer is that it probably cannot, and that the more useful framing is symbiosis — a collaborative version between human and artificial intelligence that produces something better than either alone. This is the frame he brings to the rest of the conversation.

The IQ scaling problem — and the bandwidth bottleneck: #

One of the more analytically interesting sections in the episode is Jerry’s framing of AI capability using IQ as a rough proxy. Human average is 90–100. Genius territory is 100–140. The most extreme documented human cases reach into the 200s. Current AI models, per Jerry’s cited research, are testing at 114–150. Every new model raises the number. His question is direct: what happens when AI reaches 250, or 500, or 1,000? At some point, the models start operating at a level of abstraction that most humans simply cannot follow. He uses Stephen Hawking as the human analogue — a physicist so far ahead of average comprehension that most readers of his books did not fully understand them. AI is heading toward that gap, and beyond it. The related constraint Jerry identifies is bandwidth. Humans communicate in words, roughly one token per word. AI models process at speeds that make human dialogue look glacial. If AI gets much smarter and faster, the bottleneck becomes the human interface itself.

What AI should be used for: #

Jerry is explicit that his critique is not anti-AI. The problem is not the technology. It is what the technology is being aimed at. His examples of what AI should be used for are concrete. Nuclear fusion — where OpenAI has invested in fusion research and where recent breakthroughs at Boston-based labs have moved the field closer to viable clean energy. AlphaFold — Demis Hassabis’s protein folding model at DeepMind, released open source, has enabled medical research at a global scale and helped earn Hassabis a Nobel Prize in Chemistry. Jerry cites this as the model of responsible AI: build the capability, prove the value, release it to humanity. If AI is used to cure cancer, solve Alzheimer’s, and reduce fossil fuel dependency, public sentiment shifts. Public sentiment is currently the opposite, and for reasons the industry should not dismiss.

8 billionaires deciding the fate of 8 billion people: #

The most quotable line of the episode also happens to be the substantively sharpest one. Jerry names the structural problem directly. Roughly eight people — Sam Altman, Elon Musk, and a handful of others — are effectively deciding the future direction of AI for the entire planet. Their argument, repeated across every interview, is that if we do not build AGI and superintelligence, someone else will. Jerry’s counter is that in this framing, the AI race has no finish line. It just continues infinitely. And at the end, the race is not America vs China, or Anthropic vs OpenAI. It is AI vs AI. That is the endpoint the current trajectory produces. And it is the endpoint no one at the billionaire level appears to be planning around.

The economics don’t add up: #

The economic argument in the episode is where Jerry lands the hardest. For every $1 that Anthropic or OpenAI takes in as revenue, they are spending roughly $1.50 producing it. The unit economics are inverted. On the corporate side, Anthropic is focused on enterprise customers who are burning through tokens at a rate that has already broken multiple client budgets — Jerry cites Uber having burned through their entire 2026 AI budget within the first four months of the year. On the consumer side, OpenAI has around 800 million daily users, most of whom are paying $20 a month and using the models for cheesecake recipes and email drafts. Neither business model closes the gap. The bigger, faster models being released monthly do not solve the economics. They compound them.

The bubble, the bust, and who survives: #

Jerry’s prediction is that this ends in a credit squeeze. As protest against AI grows — data centres being blocked, employees being laid off, students entering a job market with no positions — investor patience with unprofitable businesses at trillion-dollar valuations will narrow. The bubble bursts. Small players get cleaned out. Google, SpaceX, and probably Anthropic survive the reset. The AI infrastructure that survives will look meaningfully different from what has been built to reach this point. Jerry positions this not as prediction from the sidelines but as a pattern that has played out in every previous tech cycle — one that current AI investors are betting will not repeat, without a clear reason for why.

Europe is quietly building sovereign AI: #

A political undercurrent Jerry brings into the conversation is that Europe has largely made up its mind. After a series of moves by the current US administration around AI export controls and national security restrictions, European institutions have started actively distancing themselves from American AI dependencies. Jerry mentions the removal of Google, Palantir, and other American providers from European infrastructure decisions, and a broader push across the 27 EU member states — with the UK, Japan, and Singapore watching — toward sovereign AI capability. Own tech stack. Own models. Own LLMs. The strategic argument is that dependency on a foreign country’s AI is a national security problem the same way dependency on a foreign country’s semiconductors turned out to be. The financial consequence for American AI companies is significant, and largely underweighted in current valuations.

The MAD framework: what regulation should look like: #

Jerry’s proposal for AI regulation is drawn from Cold War history. During the East-West standoff, the US and Soviet Union sat down and agreed to Mutually Assured Destruction — MAD — as the framework that stopped nuclear weapons proliferation. Each side agreed there was enough capacity to destroy the world many times over, and further weapons production offered no advantage. The Paris Agreement on climate change extended the same collaborative principle to a different existential risk. Jerry argues AI needs the same table. UN-level. Governments, not billionaires, in the decision-making seat. Big tech included to advise, but not to decide. The point is not to slow AI development to zero. It is to establish rules of engagement — safe laboratory testing, guardrails before release, accountability for what models can do — that make the current wild-west incentives sustainable.

Anthropic’s responsible-scaling moment: #

Jerry highlights one specific example as the type of behaviour the industry needs more of. Anthropic reportedly developed an internal model that turned out to be too capable — able to discover vulnerabilities, jailbreak other systems, and function as an effective cyber-offensive tool. Rather than release it, they stopped it, added guardrails, and only shipped when the safety layer was in place. Jerry’s point is not that Anthropic is uniquely virtuous. His point is that the incentive structure most AI companies operate under does not reward this behaviour. The regulatory framework should. This is the connection between the MAD analogy and the practical policy path: mandatory safety testing, third-party review, and consequences for shipping models that fail either bar.

The generational close: what future are we giving our children? #

The final section of the episode is the most personally weighted. Jerry references the 142,000 tech layoffs of the past year — a figure that maps to public tracker data — and the growing number of students graduating with no job pipeline. His concern is not primarily economic. It is psychological. Mental wellbeing collapses when purpose collapses. Depression, loneliness, and self-harm risk all rise when a generation trained for four or more years is told to sit at home and collect UBI. Jerry’s closing framing is a question he keeps asking in Parliament. What future are we giving our children? Nobody, he reports, can answer it. That is the problem. The technology conversation, whatever its economics or its geopolitics, has to answer that question — or the political backlash that follows will eventually answer it in ways the industry will like even less.

Why listen: #

This episode is a rare thing in the current AI conversation — a grounded, numerically specific, willing-to-name-names critique that avoids both hype and doomsday. For founders, product leaders, investors, and anyone building or funding in the AI-adjacent space, Jerry’s framing is directly useful for calibration. The economic argument alone — the $1-in, $1.50-out math applied to trillion-dollar valuations — is the kind of thing that should be reset against every AI investment thesis on the current market. For a broader audience, the closing question is the more important one. AI is not going to slow itself down. Human institutions have to. And that means people who understand both the technology and its consequences need to be much louder than they are.