Who's your Data? Podcast

There are genuine moments of non-obvious thinking - particularly the behavioural proxy method for identifying closeted LGBTQ patients in a 5M-person database, and the harm-reduction AI framing for drug safety. However, large stretches are padded with anecdote, general AI optimism, and off-topic banter about poppers and crystal meth, dragging the density down meaningfully.

“out of 5 million patients, suddenly we went to a subset of 8,888 people. That was the number... 70 people in Clalit got infected with monkeypox. 69 out of 70 were part of the 8888 cohort”

“how do you make those AI models creating those right disclaimers that will answer like I just answered you right now instead of just wiping it out”

The behavioural-proxy approach to surfacing a hidden population in a closed medical record system is a genuinely fresh and counterintuitive idea, and the harm-reduction angle on bypassing AI safety guardrails is an underexplored framing. The Medicine 1.0/2.0/3.0 scaffold is explicitly borrowed from Peter Attia and not reworked meaningfully.

“if he's doing swabs for rectal and pharyngeal, most probably, and he's a man, most probably is gay. If he's on prep, he's gay. If he did like a vaccine for human papillomavirus from the age of 27 to 45, this is the only indication only for men”

“if you can ask the AI model is it okay and will tell you, dude, this is a very life threatening thing and you should believe me, avoid it. Instead of saying I'm sorry, I'm not allowed to answer it. This is a life saving matter”

The guest is a genuine practitioner with unusual credentials - first LGBT medicine fellowship at Mount Sinai (2020), clinical director at Israel's largest HMO covering 5M patients, and Chairman of the Israeli LGBTQ Medical Association - and speaks from real operational experience including a concrete Clalit/Microsoft project, not from a thought-leadership perch.

“in 2020, I've actually did the first LGBT medicine fellowship at the Mount Sinai in New York. This was the first time that this kind of a fellowship started there”

“I'm sitting in the main administrations of Clalit, which is the main HMO in Israel. Overseas, 5 million patients have more than half of the Israeli population”

The Monkeypox cohort case (5M patients → 8,888 identified, 69/70 infections confirmed) is unusually precise for a podcast and anchors the episode's credibility. Beyond that single set-piece, most claims about AI, adoption, and bias rest on vague appeals to 'studies' without citations or numbers.

“out of 5 million patients, suddenly we went to a subset of 8,888 people. That was the number... 70 people in Clalit got infected with monkeypox. 69 out of 70 were part of the 8888 cohort”

“most of the studies that are repeating on and on are always talking about 8 to 11% of the population is part of the LGBT, uh, community”

The host is personally invested and asks broadly sensible questions, but frequently hijacks the floor with extended personal anecdotes and opinions (the ChatGPT/13-doctors story, South Carolina ER, poppers commentary) rather than probing the guest's expertise. There are no real follow-up challenges or pushback on any claim.

“And I will say that also the gays we know, we can always tell you what divas are up and coming. So always listen to us”

“I think that it's problematic in the way that it's presented because it's not right. You have to really think about what this means and what this doesn't mean. It doesn't mean Chat GPT is a better doctor”