‾‾‾|‾‾‾-shaped Innovator | Health & Tech | 🏃🏻 Workshops, hackathons, rapid prototyping 🚀 powered by generative AI & automation
Antwerp
Joined June 2008
- Tweets53.1K
- Following1.4K
- Followers5.7K
- Likes17.8K
Pinned Tweet
18 platforms checked against their own terms. None of the AI app-builders sign a BAA.
Lovable, Bolt.new, Base44, Replit and v0 all sit at not-applicable: a BAA is a property of the infrastructure holding the data, not the tool writing the code. The layer underneath signs. Firebase and the big three cloud model APIs by default; Vercel, Supabase, OpenAI and Anthropic on a paid tier; Netlify, Mistral, Clerk and Resend at Enterprise only.
No BAA means no PHI on that vendor, on any plan, at any price. Every row carries a verified date and a watcher that re-checks for drift.
Which row surprised you?
cliniciansthatcode.com/guide…
Bart Collet retweeted
Un salaire proche de 16.600 euros par mois : le président de la FGTB refuse de dévoiler sa rémunération exacte 21news.be/un-salaire-proche-…
Bart Collet retweeted
Nutella is 13% hazelnuts. The first two ingredients are sugar and palm oil and together they're more than half the jar. Ferrero doesn't hide it, the percentage is printed on the label in Europe.
Ferrero buys around a quarter of the world's hazelnut crop every year, mostly from Turkey, which grows about 70% of the supply. In 2014 they bought Turkey's biggest hazelnut company so nobody could get ahead of them.
In 2017 they changed the recipe to more sugar and more milk powder without announcing it, and a German consumer group noticed the cocoa went down too because the color got lighter.
Twice the diagnostic accuracy of generalist doctors on hard cases is the headline. The more interesting part is the design.
MAI-DxO is not one model. It is several agents acting as a panel: asking questions, ordering tests, debating the diagnosis. Results varied by underlying model, so orchestration matters.
Still a simulation, though. The Boston trial alongside real clinicians is the test that counts.
If the panel is where the gain sits, who owns that layer: model labs or hospitals?
techcrunch.com/2026/10/22/mi…
One AI tool lifts productivity. Two, more. Three, still more. Then it drops: you are no longer delegating, you are conducting.
People overseeing multiple agents report 14% more mental effort, 12% more fatigue and 33% higher decision fatigue. Researchers call it AI brain fry; I call it the conductor problem.
I once ran five agents in parallel. Two hours in I was exhausted from overseeing, not doing. You did not get a delegation tool; you got a new job title. So did the radiologist reviewing twenty AI-flagged scans.
How many agents before your oversight slips?
hyperadvancer.com/blog/the-c…
Bart Collet retweeted
De jongerenpartijen van de meerderheid doen een oproep naar meer verantwoordelijkheidszin. De @JSbelgie hebben dit niet ondertekend.
Bart Collet retweeted
CALLING ALL BUILDERS! 📢
The future of healthcare and biology is being built with AI. Come build it with us.
We're bringing ~100 exceptional engineers, founders & hackers together for the Health Frontier NYC Hackathon, co-hosted with @generalcatalyst & @percepta!
1.5 days to build, experiment & ship. Healthcare experience not required.
Huge thanks to our full list of partners:
@hippocraticai @modal @DecagonAI @judihealth @poweredbyfabric @cognition @certifyos @YuzuHealthInc
📍 NYC | Dec 4–5
👉 Apply by 11/24 (link below & limited spots!)
310 seconds: the average time to deploy a governed, private clinical AI service in a tertiary care centre (n=25).
The bottleneck was never the model. It was the bespoke setup of secure backends, private endpoints and access rights, rebuilt hospital by hospital.
Infrastructure as Code turns that into one template, reviewed once and deployed identically anywhere. The catch: identical deployment also replicates identical mistakes. Who signs off the template, and how often?
🚨blood and plasma refrigerators 🚨
Cyberattack on France’s national blood service (EFS) via its refrigeration contractor Coolworld. Hackers reset freezers keeping blood donor samples at -30°C to a target of 110°C. The thing that stopped it from destructive impact was an on-call technician who noticed the temperatures going up. Frozen plasma intended for transfusion was stored there too. Fortunately nothing was destroyed. This was a cyberattack on a sensitive layer of healthcare. From frozen peas to boiling plasma in three weeks...?
💯 Het echte probleem is dus niet dat de sterkste schouders te weinig belastingen betalen, integendeel, maar dat er veel te weinig sterke schouders zijn, en dat er veel te veel mensen afhankelijk zijn van die veel te kleine groep van sterke schouders. 💯
De “sterkste schouders” dragen al de zwaarste last
Het politieke mantra dat “de sterkste schouders nóg meer moeten bijdragen” klinkt redelijk, maar houdt na analyse weinig stand. De cijfers tonen dat de hogere inkomensgroepen al het overgrote deel van de personenbelasting betalen, en dat België Europees koploper is in de belastingdruk op arbeid. Nog verder belasten is daarom onrechtvaardig én onhoudbaar.
De cijfers van de Statbel-fiscale statistieken maken de progressiviteit van het systeem duidelijk:
- De onderste 50 procent betaalt naar schatting slechts 7 procent van de personenbelasting.
- De middelste 25 procent draagt zo’n 23 procent bij.
- De top 25 procent staat in voor zo’n 70 procent.
De bovenste helft financiert het overgrote deel van de personenbelasting, terwijl lage inkomens grotendeels onder de belastingvrije som of in de lagere schijven blijven.
België is bovendien structureel Europees kampioen in de belastingdruk op arbeid. Volgens het OESO-rapport Taxing Wages bedroeg de tax wedge (het verschil tussen totale loonkost voor de werkgever en het netto loon voor de werknemer) voor een alleenstaande gemiddelde werknemer zonder kinderen ongeveer 52,6 procent in 2024, de hoogste van de hele OESO, ruim boven het gemiddelde van circa 35 procent. Het topmarginale tarief (rond 50-53,5 procent inclusief opcentiemen) wordt al relatief vroeg bereikt, vanaf een belastbaar inkomen van omstreeks 50.000 euro, veel sneller dan in Duitsland of Frankrijk. Ook de totale belastingdruk (tax-to-GDP) behoort tot de hoogste van Europa, met ratio’s van 44-45 procent van het bbp.
De huidige verdeling is dus al enórm progressief in België. De groep die het meeste betaalt, ontvangt relatief weinig terug via transfers. Hoge marginale tarieven ontmoedigen elke extra inspanning, investeringen en ondernemerschap, en verergeren de al (veel te) lage activiteitsgraad.
Het echte probleem is dus niet dat de sterkste schouders te weinig belastingen betalen, integendeel, maar dat er veel te weinig sterke schouders zijn, en dat er veel te veel mensen afhankelijk zijn van die veel te kleine groep van sterke schouders.
Oplossing? Meer mensen aan het werk in de private sector en een efficiëntere overheid leveren veel meer op dan nog een extra belasting. Alleen dát is een duurzame en structurele lange termijn oplossing voor het failliete België. Nog eens extra belastingen voor de groep die al het leeuwendeel financiert, negeert de bestaande realiteit en ondergraaft de draagkracht die onze welvaartsstaat nodig heeft.
Longevity does not spread its big weeks evenly. This week the circuit stacks four flagship gatherings: Longevity Week across London, ICSA in South Korea, the GSA meeting outside Washington, and the Eurosymposium on Healthy Ageing in Brussels.
That is the shape of the calendar. We track 500+ longevity events across 40+ countries, and they cluster into a few hubs and a few weeks. London and Berlin carry Europe. The US splits East and West.
Planning around density beats chasing single events.
Which cluster are you travelling to this quarter?
longevents.hyperadvancer.com…
57% of frontline highway crews call placing traffic cones their most dangerous task. The US recorded 850 work zone deaths in 2024.
SkyCone is one of the smartest drone adaptations I've seen. The drone doesn't carry the cone; it is the cone. It flies to the incident, lands, and folds its four arms into the striped pyramid.
Every cone machine so far still needs a truck crawling along live traffic. Here each cone delivers itself. Still a student concept, though. What else should deliver itself?
yankodesign.com/2026/11/02/f…
Clinicians live with the IT problems every day. Now they can build the fixes. #cliniciansthatcode
Start with the friction clinicians actually experience. On @NEJM_AI Grand Rounds, Dr. @mnedoszytko says living inside the problem can reveal what needs fixing — and why physicians may be well positioned to build a useful solution for care. Full episode: nejm.ai/ep47
Are you building the next big thing? SXSW Pitch applications close in 10 days, on Friday 13 November.
If you're a founder with an idea ready for a bigger stage, this is your invitation to apply. Finalists pitch to a panel of industry experts, early adopters and investors in Austin on 16-17 March 2027.
Healthcare founders: there is a dedicated Life Sciences, Healthcare & AssistiveTech category. Eligibility: launched after 1 January 2024 and under $10M raised.
Know a startup we should see? Tag them or share this post with them.
cart.sxsw.com/products/accel…
25,000 Dutch central government jobs will go by 2035, about 16% of the workforce. The headline says AI. The letter to parliament says something more useful.
The target is €900 million in savings in 2027, €3.5 billion a year from 2035. Next to AI: scrap or simplify 500+ rules a year, trim the advisory bodies, standardise IT, review every departure instead of freezing hiring.
"Simpler, smarter and smaller", and the order matters. AI pays off on work that was simplified first. What would your government simplify before it automates?
qz.com/netherlands-cut-25000…
question for @herdrdev and @NousResearch : How can i keep track of the kanban progress in Herdr? Can't get a decent result apart from a small script that i've open in a tab.
Bart Collet retweeted
Ozempic and Mounjaro beat Metformin at preventing prediabetes from progressing to diabetes, and they also seem to protect the heart.
In health records from 256,492 people with prediabetes and obesity, people prescribed a GLP-1 drug had 31% less type 2 diabetes over about 3 years than those prescribed Metformin, along with 15% fewer heart attacks and 20% fewer strokes.
Metformin still has solid evidence behind it. In 2002, it cut new diabetes by 31% compared with a placebo.
The death number is the one to treat carefully. Deaths were 41% lower here, but Wegovy, a higher dose of Ozempic, cut heart events by 20% in a trial of 17,604 people. That gap probably reflects who gets these drugs, not a clean head-to-head result.
The diabetes finding is backed by trial data. Over 3 years on Mounjaro, 1.3% of people with prediabetes developed diabetes, versus 13.3% on placebo.
Bottom line: if your blood sugar is creeping up, ask your doctor whether a GLP-1 drug or Metformin makes sense for you.
