SmarterMarkets™

There are a handful of genuinely useful frames - the factor-of-10 cost cascade, the hidden human-in-the-loop data-fix role embedded in every job, and the creator/customer duality - but they are surrounded by substantial repetition and general platitudes that dilute density. Many ideas are stated but not developed with enough depth to be immediately actionable.

“companies hire these high price skilled with connection salespeople and then a third or a half of their job is dealing with data issues”

“if you can deal with an issue here, and it cost a bone, right? A euro, a pound, a dollar, whatever it is, and instead you deal with it downstream, you it's going to cost 10”

The 'AI is data' slogan and the factory-vs-lab argument for AI deployment are moderately fresh framings, and the student-gaming-the-LLM example is a clever customer-focus illustration. However, most of the underlying ideas (data quality as fitness for use, start with the customer, data definitions not a tech problem) are standard quality-movement doctrine that Deming-era practitioners would recognise immediately.

“too much AI is being done in the metaphorical factory...AI uh is the complete opposite of that. It is to get things out of control, to get them to a new level”

“we'll have frat boy confidence in our answers and whatever happens, happens. And what's happened so far is slop”

Tom Redman is a genuine long-tenure practitioner - Bell Labs data quality lab from the late 1980s, decades of advisory work inside real companies - which gives him legitimate credibility as a doer rather than a pure thought-leader. He is not, however, a widely-recognisable name who operated at organisational scale as a senior executive, and he occasionally drifts into consultant-generalism.

“in the late 80s and early 90s, we set up a little lab. It's a group of guys to explore data quality”

“I've, I've been through, uh, advised on or one way or another, been connected on 14 things that I involve that I include as transformation”

A few concrete anchors exist - the Google researchers' paper title, the attributed OpenAI quote, the Nobel Prize protein-folding example, and the factor-of-10 rule - but no named client case studies, no revenue or error-rate data, and the factor-of-10 is asserted without sourcing. The episode leans heavily on illustrative analogies rather than verifiable evidence.

“there's a great paper by Google researchers called everybody wants to do the model work, Nobody wants to do the data work”

“a great quote. I think it's from a guy named James Beckter at OpenAI. The IT in AI is data”

The host connects ideas reasonably well and occasionally surfaces a genuinely useful bridge (e.g., linking agents back to the creator/customer frame), but there is no meaningful pushback, no probing of the factor-of-10 claim for evidence, and several questions are leading or confirmatory rather than challenging. The interview stays safely within territory Redman is comfortable in.

“And there's been a move to push more towards, like, these large pools of data that are available. When you bring up that example, is the answer there to go to a larger, more flexible pool of data, or is it just to understand that you're going to have different sets of data for different needs”

“Oh my goodness, I'm so glad you asked that question”