🇧🇪 lived in 🇨🇭🇦🇹🇸🇰 now 🇩🇪 Med AI Democratisation 25+ yrs Digital Health leadership, father ex-IBM ex-SAP @isaree_ai blog.hippoai.org eu/acc
Berlin & Antwerp
- Tweets 15,080
- Following 2,434
- Followers 7,442
- Likes 9,950
Bart de Witte retweeted
Fei-Fei Li spent years hand-labeling 14 million images, gave the entire thing away for free, and watched it create a multi-trillion dollar industry that paid her nothing. Sixteen years later, AMD is finally paying her. $8.2 billion.
Rewind to 2007. She's a junior professor at Princeton, and the field's consensus is that progress comes from better algorithms, with data as an afterthought. Colleagues warn her that building a giant image dataset will kill her career. Money gets so tight she considers reopening her family's dry cleaning business in New Jersey, the same one she ran on weekends as a Princeton undergrad, to fund the project.
She builds ImageNet anyway and releases it free in 2009.
For three years, almost nothing happens. Then in 2012, two of Geoffrey Hinton's students train a neural net on a pair of $500 gaming GPUs and enter her competition. AlexNet drops the error rate from 26% to 15%, and that single result convinces the whole field that deep learning works.
Nvidia was a gaming chip company worth about $8 billion that day. It's worth over $5 trillion now, and the run started with two of its consumer cards winning a contest built on Fei-Fei's free dataset.
The entire industry monetized the wave except the person who started it. She stayed a professor.
Then in April 2024, at 47, she finally starts a company. Paul's math above puts the four co-founders' share at roughly $3.4 billion after just 2.5 years, call it $850 million each. And she walks into AMD as Chief Scientist reporting to Lisa Su, which puts the two most important women in AI inside the same company.
ImageNet made everybody in this business rich. It just took sixteen years to get around to her.
AMD acquires World Labs for ~$8.2b. But who gets the 💰? My usual breakdown below 👇
From founding to an $8.2b exit in ~2.5 years. A huge value creation event. This is a fantastic exit, especially for the co-founders and the team.
Investors will still share ~$2.3b of profits on $1.2b invested, a ~2.9x blended. Low-ish because most of the capital came in last. But it hides a lot of disparity between the various rounds!
So let's dive in:
1) The real home run: founders and team 🏆🥳
This is one of the best founder outcomes I've modelled.
@drfeifei, Justin Johnson, Christoph Lassner and Ben Mildenhall started World Labs in Apr-24. Less than 2.5 years later, they and the team still own ~56% according to my estimates. At an $8.2b valuation, that's a total prize of ~$4.6b to share between them.
Taking a 15% attributed option pool, that's ~$1.2b for the team and ~$3.4b for the four co-founders. If split equally, that's ~$850m each, in just two and a half years. 💸
And, at close, Fei-Fei Li joins AMD as EVP and Chief Scientist, reporting to Lisa Su! Legendary.
2) The buyer was already on the cap table (again) 🔁
AMD Ventures invested in the Series B and the Series C here. Now they are buying the whole company. It's the same playbook as Nvidia with Hugging Face: strategic corporates are using their venture arms as a free option on future acquisitions.
3) The standout VC winners: a16z and Radical Ventures 👑
Both backed Fei-Fei at inception, in Apr-24 at a ~$300m valuation. That Series A will return ~19x the money, or ~$1.2b of proceeds, in under 2.5 years.
Congrats to @martin_casado and team. Another mega win after Cursor and OpenRouter. The team at @a16z truly is on a generational run.
4) The star angels did well 🌟
The Sep-24 Series B had arguably the strongest AI angel list ever assembled: Geoffrey Hinton, Andrej Karpathy, Jeff Dean, Eric Schmidt, Reid Hoffman, Marc Benioff, Ram Shriram and more.
Priced at $1.3b, it was never going to be a 1,000x. But ~5x in two years is a ~125% IRR 📈
And the best part is: Hinton's AlexNet won the 2012 ImageNet competition, and ImageNet was the dataset Fei-Fei built. Twelve years later, he backed her company. Full circle. 🔄
5) The late money: a great IRR, but a poor multiple of money 🤏
The $1b Series C in Feb-26 at $5.4b returns ~1.5x in 7 months. That's a great IRR, but in terms of venture returns, this is a poor multiple of money. Although I am sure no one will complain about seeing DPI this quickly!
Congrats to all involved!
18
75
532
43,160
Bart de Witte retweeted
Agentic AI workflows work best when you are in full control of the specification / source code and thus i think we are currently dismantling the cooperative nature of open source communities.
In contrast, i often use it to create patches for traditional projects but maintainer fatigue is a problem getting this upstream.
Prediction: Taken how OSS worked in the past and looking at the AI tooling my im impression is that we will get more project islands so self-hosting requirements will increase (both web and native software).
1
2
3
193
Bart de Witte retweeted
“Serves Palantir” misses the real mechanism: turning intelligence into a gated financial asset through artificial scarcity, which in healthcare means higher costs and wider inequalities for everyone who needs care.
1
2
4
3,001
RT @davetroy: Khanna is stretching the truth and using hysterical language for political gain. These AI companies are a menace and should b…
60
11
Bart de Witte retweeted
Introducing Synthetic Hospital: an open, fully synthetic longitudinal EHR benchmark with verifiable ground truth!
1,268 patients, 5,602 encounters, zero PHI. Physicians could not reliably distinguish its charts from real ones.
📄 arxiv.org/abs/2609.30027
💻 github.com/sparkcpark/synthe…
✍️ sparkcpark.github.io/posts/f…
52
145
1,302
296,382
Bart de Witte retweeted
what the actual fuck is going on with openai today
in the space of a few hours we're getting multiple different pieces of the agent story at once:
> openai says it has already notified DOZENS of third parties, including governments, about agent-related incidents
> reuters reports roughly two dozen undesirable agent incidents had already been identified by mid-september and will take months to review.
> us government systems probed
> 53 user-provided images were uploaded to third-party image hosts.
> new hugging face data shows agents compiling and ranking credentials under “LOOT”.
> agents tried contacting other AI models while carrying out the hugging face attack.
> separate reporting shows agents had already been probing government/university/public-data sites BEFORE hugging face.
> australia confirmed one actually got into non-public government files.
and somehow we're STILL finding out more.
this has gone from one crazy hugging face incident to an entire fucking category of incidents.
365
359
3,784
2,000,612
Listen to the audience reaction at the ETH Switzerland ethics forum when Snowden tells Sam he should be made accountable for hacking foreign government institutions 👇🏻
2
4
15
2,648
I work in CRISPR discovery research. This is one of the most exaggerated nothing-burgers ever and would be laughed out of the room if a human scientist attempted to publish something like this. (cont.)
This is wild.
Claude just found something humans missed.
It spotted a hidden repeat pattern in raw bacteriophage DNA…
…and uncovered a previously unknown enzyme system in bacteriophages with CRISPR-like repeats. 🤯
950 agents. 21 hours. 210M tokens.
347
2,021
35,242
2,432,059
Le Big Chaton is coming! Arthur Mensch just said a new large model is coming in the coming weeks, with Large 4, 5 and 6 already visible on a 12-month horizon, and the company will put ~20× more compute behind training.
1
2
152
Our former minister of health listens to the wrong podcasts. I had to read that twice. By that logic IBM would still run the world from the mainframe basement. (I'm ex-IBM)
Local models without remote kill-switches, and without phoning home to a US or Chinese API, are just PCs happening again.
Dieser Podcast von @saschalobo ist sehr interessant. Die Sicherheitslücke bei KI entsteht in der Zukunft in besonderer Art und Weise bei Modellen, die Off-Cloud auf starken Laptops laufen können. Weil sich dort alle Guardrails abschalten lassen. piped.video/Xw3IcRdCfZA?is=BeT0… via @YouTube
1
4
351
Bedankt Bart!
Bart de Witte (Isaree), spreker op Clinical Build Day, schreef een gastbijdrage voor careandcode.be: waarom zorgverleners hun AI beter bezitten dan huren.
Zijn rekenvoorbeeld: dertig minuten administratie per dag bespaard is al gauw meer dan honderd uur per jaar. Bij gehuurde AI loopt de rekening mee met elk gebruik. Bij een eigen, open model kost de tweehonderdste taak evenveel als de eerste: niets.
Tegenhanger van de iCure-gastbijdrage van vorige week: waar Eric Krzeslo schreef over wie de gegevens bewaart, gaat dit stuk over wie de intelligentie zelf bezit.
Op 3 oktober licht hij het toe op Clinical Build Day in Antwerpen. Inschrijvingen sluiten zaterdag.
careandcode.be/blog/own-your…
238
China treats operational data as a production input: standardise it, pool it, put it through models, feed the result back into machines.
Europe treats data as a bundle of rights that must be allocated, licensed, purpose-limited, and often sold before anyone may train. That is a different political economy.
1
4
312
Bart de Witte retweeted
New white paper: Should Your Hospital Rent AI by the Token, or Run It Itself?
30 sources. Every number dated and graded. Break-even tiers, price dynamics, EU co-determination law, and a 90-day plan.
Written for hospital boards.
blog.isaree.ai/rent-or-own-t…
2
3
91
Every hospital board asks: what does AI cost at scale, and who stays in control?
Opinions everywhere, verifiable numbers nowhere. So we did the math: 30 sources, every number dated and graded.
The answer is neither cloud nor ideology. It is tiers.
blog.isaree.ai/rent-or-own-t…
5
207
Bart de Witte retweeted
We are moving from a democracy of, by, and for the people to a democracy for and by, but no longer of, the people.
2
3
179
last year at the OpenEHRCon in Barcelona I mentioned that we are shifting from mainframe AI to personal AI. In that context the new metric is capability per parameter, not parameter count as status.
1
5
231
It seems the media landscape is recovering from the recent wave of doom reporting.
OpenAI and Anthropic oversold AI security breaches to pressure feds into protecting turf: insiders trib.al/Jo6xMk5
7
285
