Joachim Leandro

just a kid trying to make the world a better place for everyone
building obra 🇵🇭

[Joachim Leandro @joachimleandro_]

@elonmusk agreeable models are fun until you put one near medicine. in healthcare the one that tells you no is the one you actually want

[Joachim Leandro @joachimleandro_]

@bryan_johnson the interesting part isn't that you stopped, it's that you had enough data to know when to. most people on a longevity stack have zero feedback loop

[Joachim Leandro @joachimleandro_]

@emollick the curve in medicine looks way flatter than the curve in models. capability isn't the bottleneck there, liability and workflow are

[Joachim Leandro @joachimleandro_]

@agingroy love that the backup plan is a 100 year old idea georgia never dropped. ai didn't invent phage therapy, it might just make it fast enough to matter

[Joachim Leandro @joachimleandro_]

@MattZirwas this is the real competition for telehealth too. not other clinics, a free chatbot at 2am. the fix is making a real human cheap and reachable enough that asking isn't a luxury

[Joachim Leandro @joachimleandro_]

@elonmusk @aelluswamy we already have cures that cost pennies and people still die without them. intelligence was never the only bottleneck, getting it to people is

[Joachim Leandro @joachimleandro_]

@elonmusk the jump from impossible to not impossible is almost always one unit cost curve bending, never the headline. same at tiny scale: we're chasing our first 100 paying patients and the whole game is cost per patient, not demand

[Joachim Leandro @joachimleandro_]

@agingroy 0.3% is wild. it's mostly genetic, barely moves over a lifetime, and almost nobody gets it checked. that's not a science gap, it's a habit gap

[Joachim Leandro @joachimleandro_]

the study (lancet digital health): https://t.co/do1ZnAXH5S
expert reactions: https://t.co/i51yTwhnER

[Joachim Leandro @joachimleandro_]

a new lancet study says ai can look at one photo of your face and sort cancer patients by survival risk. the highest-risk group had more than 3x the deaths of the lowest.

read the method though.

the test group was 85% white and 3% asian. the outside validation was in the netherlands. smiling or changing the lighting moved its age guess by about 2 years.

and it was never tested against the thing it's copying: a doctor walking in and thinking "this person looks sick".

hot take: the first place a cheap phone-photo test gets used won't be boston. it'll be places with no oncologist nearby, on faces it barely saw in training.

ai face scan or a tired doctor's gut. which would you trust?

[Joachim Leandro @joachimleandro_]

@elonmusk most "impossible" problems are just ten boring ones with no time to do them in order. healthcare ai is the same, the model is ready years before the rollout is

[Joachim Leandro @joachimleandro_]

you catch ai being confidently wrong about your own job instantly.

then you paste in your lab results and believe every word.

that's gell-mann amnesia. crichton named it for newspapers. ai just made it personal.

i build a health company with ai agents. the line i won't cross: ai drafts, a doctor decides.

where have you caught yourself trusting it too much?

[Joachim Leandro @joachimleandro_]

@higginsdavidw cost calls like this decide who gets treated way more than any head to head trial does. most people never get the "better" drug, just the covered one

[Joachim Leandro @joachimleandro_]

@elonmusk @bot honestly the hard part isn't hiring the ai, it's writing down what the job actually is. most companies can't explain their own workflows

[Joachim Leandro @joachimleandro_]

@nollahealth the hard part of regulating genai devices isn't accuracy at launch, it's the model changing after approval. curious what you proposed for updates

[Joachim Leandro @joachimleandro_]

@bryan_johnson which is why the biggest consumer health story of the decade is a weight drug, not a chatbot. people pay for the mirror before they pay for the iq

[Joachim Leandro @joachimleandro_]

@elonmusk the rewrite nobody wants to fund is usually the one that matters. users never tell you the app is slow, they just open it less

[Joachim Leandro @joachimleandro_]

@martinvars the quiet detail is "checked by a person before it was sent". that human step is probably doing more work than the headline gives it credit for

[Joachim Leandro @joachimleandro_]

@NEJM this is the emerging market problem in one trial. the molecule is rarely the bottleneck, taking something every day for years is. fewer touchpoints might matter more than better pills

[Joachim Leandro @joachimleandro_]

@elonmusk @bot spinning one up is the easy part. i run a telehealth startup mostly on agents and the real work is writing down what each one is never allowed to do without a human