AI Researcher With 20 Years Experience Cuts Through Hype

The AI conversation has reached a curious pitch. Everyone has an opinion, most people have an agenda, and the signal-to-noise ratio approaches zero. Which makes Dimitri—a researcher who's been working in AI for over twenty years—worth listening to.

He appeared on The PrimeTime's podcast not to sell anything or predict the Singularity, but because his non-technical friends keep asking him questions. Lawyers, accountants, doctors, parents with kids in college. They want to know: Is this real? Should I worry? What happens next?

"I wanted to know what I was talking about before I said something," Dimitri explained about starting the podcast with Casey. "Which puts him in the top 20% of podcasters at least."

Fair point.

The Question Nobody Can Answer

The hosts pressed Dimitri on token costs—specifically Sam Altman's claim that AI will become 100x cheaper within two years. It's the kind of promise that sounds either visionary or absurd depending on your proximity to the technology.

Dimitri's answer: "I don't know."

Not a hedge. Not false modesty. Genuine uncertainty from someone who shipped his first custom AI design at Google in 2005, back when the field was quiet and nobody expected this stuff to work.

"A lot of that is infrastructure development and algorithm development that is trade secrets," he said. "Even an OpenAI insider, it would be illegal for them to evaluate in public."

But he offered something more useful than a prediction: a framework. "The timing matters as much as the content of the claim," Dimitri noted, pointing to Elon Musk's history of promises. Reusable rockets? Musk called it in 2005. It worked—eventually. Full self-driving in Teslas? Promised in 2016, still waiting.

"I would not be surprised if eventually we can get 100x cheaper token cost," Dimitri said. "Whether or not they can do it now, that's beyond my knowledge."

The Land Rush Problem

What Dimitri does know: the current AI landscape resembles a land grab. Multiple big players racing to build whatever they can, as fast as they can. Ship it if it works. Ship it if it barely works. Move to the next thing.

"Some of that is maybe recklessness, but some of that is just market forces," he said. "Right now there are multiple really big players and probably there won't be as many big players in 10 years. Given how much money they've all put into this already, they really don't want to be the ones caught out."

The arms race makes business sense even when it produces questionable engineering. But it creates a specific problem: instability. If you spend a year optimizing your stack to reduce token costs, then someone introduces the next architectural innovation, does your optimization become worthless?

"My biggest question would be: is the internal stack stable enough that you can optimize it now?" Dimitri said.

Nobody knows. Not even people deep inside these companies.

Google: The Quiet Monster

One participant raised an interesting point about Google. If massive cost reductions were truly imminent, wouldn't Google already have them? They've been building AI-specific hardware—TPUs—since at least 2014. They have infrastructure advantages nobody else can match. They have what Dimitri delicately called "surveillance on the entire internet"—a constant flow of trainable data.

"Google is sort of the quiet monster in the business," Dimitri said. "Many people have this feeling of: possibly Google is just going to quietly leverage all those advantages that nobody else has."

Yet they've been oddly slow on the product side. The chat interfaces, the consumer-facing tools—they lagged. Which either means the cost advantages aren't there yet, or Google is playing a longer game than quarterly earnings suggest.

XAI, meanwhile, is talking about orbital data centers. Space-based AI infrastructure. Dimitri found it notable that their designs rely on conventional Nvidia GPUs, not TPUs. Either they don't think they can build or license something better, or they know something about the trade-offs that isn't public yet.

"If there's one thing that category of companies does well," Dimitri said about Musk's ventures, "it's relentless execution on building and optimizing physical stuff."

The Career Question

Beneath all the technical discussion sits a human problem. Dimitri sees junior engineers wondering if they should even start. He sees 45-year-old senior engineers—people with stable careers and good incomes—wondering what happens if the technology actually works in five years.

"That's an especially bad time to be useless, right?" Dimitri said. "If you're 50 years old and don't have 15 years of building up a new career ahead of you, what do you do?"

He doesn't have an answer for that either. Just the observation that people are asking.

This is where Dimitri's perspective matters most. He's not selling courses on prompt engineering or consulting packages on AI transformation. He's watching people he knows—lawyers, doctors, his own colleagues—try to make practical decisions with incomplete information during what might be a genuine inflection point or might be 2024's version of the blockchain.

"In the same way that you can't take whatever Microsoft claims at face value about improvements in Windows performance," Dimitri said, "you can't take AI company claims at face value either."

Twenty years in this field has taught him to separate what's technically possible from what's commercially viable from what's actually shipping. Most people talking about AI haven't made those distinctions. Dimitri has, which is why when he says "I don't know," it carries more weight than someone else's certainty.

The hype will continue. The promises will get bigger. The only question is whether the people making decisions—about careers, about businesses, about investments—will have access to grounded perspective when they need it.

—Bob Reynolds, Senior Technology Correspondent