Dapper Data

There are a handful of genuinely informative segments - particularly around NER-based script breakdown and generative AI's precision limitations - but the episode is padded heavily with introductions, music chat, host self-commentary, and a lengthy off-topic party game. The ratio of actual insight to filler is low.

“the core technology solves the um, thing called named entity recognition. So basically just that you want to distinguish different type of object in text. So you prepare data set with correct labels basically telling which, which type of what content means what”

“once we will be able to actually make machine understand a bit more about the functionality of objects uh and stuff it tries to generate, we'll put it to the next level”

Nearly every claim is a recycled mainstream take: ChatGPT hallucinates, AI won't replace humans but will be a tool, you can't stop progress, generative AI struggles with hands. There is no contrarian framing, no first-principles reasoning, and no unexpected angle on the film-tech intersection.

“it's not a library uh, of hundred percent uh proven facts. It's uh, a really complex neural network which generates response to your question basically”

“I don't Think it's going to replace like, people and uh, some professions. But it will be a really nice tool for those who want to be competitive”

Ruslan is a legitimate co-founder and CTO who built and shipped a real ML-powered SaaS product for a specific industry vertical, giving him genuine practitioner credibility. However, the company is early-stage (~4 years old, small team), he is not operating at significant scale, and his domain expertise beyond the product itself is limited.

“Three, uh, friends from the university we graduated together with Igor and Andrei back to 2012 in Minsk, Belarus”

“we started to build our uh, model, started to build the data set, we started to run tests and actually we got really nice results uh, after a few experiments and we decided, okay, so we're going to uh, hit the market with this idea”

There are some concrete anchors - 100-page scripts equating to ~100 minutes of screen time, script breakdown taking days to weeks, Google Cloud as the infrastructure, NER as the named technique, and the real Drake/Weekend AI song incident - but the episode never surfaces user numbers, accuracy benchmarks, revenue, or the kind of hard evidence a B2B operator would use to evaluate the product or the claims.

“It's a piece of text usually 100 pages long for a feature film, so which translates to something around uh, hundred minutes of screen time”

“takes anything from a couple of days to a couple of weeks or even longer depending on the complexity”

The host relies almost exclusively on wide-open, softly framed questions ('How is it changing the world?'), heaps excessive praise throughout, mispronounces the guest's name repeatedly, and devotes the final segment to a completely irrelevant 'Overrated/Underrated' game. There is no meaningful pushback, no probing follow-up, and no productive tension.

“How is it changing the world? How is it making a difference in technology right now, now?”

“On the verge of being the next Einstein man. Or close to it, man. That's what I'm talking about”