In Conversation with General Intuition's Pim De Witte: Tackling Ebola to Building Foundation Models
Why It Matters
De Witte’s blend of crisis‑tech innovation and game‑derived AI data demonstrates a scalable path for building more grounded, action‑oriented models, reshaping both humanitarian response and the future of artificial intelligence research.
Key Takeaways
- •Ran a 2‑million‑player RuneScape server, mastering self‑service scaling
- •Turned hackathon experience into Ebola tablet solution for field data
- •Developed MapWipe, crowdsourced satellite labeling for inaccessible disaster zones
- •Created Metal TV, capturing granular game actions to train AI models
- •Advocates world‑model and policy architectures over pure LLM approaches
Summary
In this interview, General Intuition CEO Pim de Witte recounts an unconventional career that began with running a massive private RuneScape server and evolved into cutting‑edge AI research. He describes how the server, which hosted up to 5,000 concurrent users and generated millions in revenue, forced him to learn security, scaling, and team‑building without any corporate safety net.
De Witte’s hackathon win caught the attention of Ebola response teams, leading him to design a chlorine‑dip tablet for field data collection in low‑connectivity environments. He later built MapWipe, a crowdsourced platform that distributes satellite imagery to volunteers, creating up‑to‑date maps for disaster zones where commercial data is scarce. These experiences fed directly into Metal TV, a video‑recording system that logs precise in‑game actions rather than keystrokes, providing a rich dataset for training general‑purpose AI models.
He argues that large language models (LLMs) are limited by their text‑only view of the world, likening them to a blind snowball rolling downhill. Instead, de Witte champions world‑model and policy architectures that learn through actions and observations, integrating spatial‑temporal dynamics missing from pure text generation. He cites Metal TV’s granular action data as proof that games can bridge digital and physical information streams, offering a new pre‑training paradigm.
The conversation underscores how hands‑on problem solving in crisis response and gaming can generate novel data pipelines that accelerate AI development. De Witte’s trajectory illustrates that unconventional, self‑directed projects can produce both humanitarian impact and foundational advances for next‑generation artificial intelligence.
Comments
Want to join the conversation?
Loading comments...