ChatGPT and Cancer: How a Tech Founder Rewrote His Treatment Plan

Key Concepts

• Founder Mode in Medicine: Applying entrepreneurial intensity, first-principles thinking, and rapid iteration to navigate a medical crisis when standard care is exhausted.
• Maximum Diagnostics: A comprehensive approach to data collection (e.g., single-cell sequencing, bulk RNA/DNA sequencing, pathology staining) to inform personalized treatment.
• AI-Augmented Research: Using AI agents to perform literature reviews, analyze complex bioinformatics data, and bridge the knowledge gap between patients and medical specialists.
• Personalized Therapeutics: Developing bespoke treatments, including mRNA vaccines, TCR-T (T-cell receptor) therapy, and CAR-T (chimeric antigen receptor) therapy, tailored to an individual’s specific tumor profile.
• Systemic Bottlenecks: Challenges in the current healthcare system, including rigid Institutional Review Boards (IRBs), high costs of clinical trials, and misaligned incentives between doctors (risk aversion) and patients (survival maximization).

1. The Medical Journey and "Founder Mode"

Sid Severy, co-founder of GitLab, was diagnosed with osteosarcoma, a rare and aggressive bone cancer. After standard treatments failed and the cancer recurred, he and geneticist Jacob Stern adopted a "founder mode" approach to his care. They bypassed traditional limitations by:

• Data Collection: Aggregating 25 terabytes of diagnostic data (available at osteiosark.com).
• Parallel Processing: Running multiple experimental treatments simultaneously to maximize survival chances.
• Single-Patient IND: Utilizing the FDA’s "Single Patient Investigational New Drug" pathway, which allows for experimental treatments for individuals when no other options exist.

2. Technical Methodologies and AI Integration

Jacob Stern detailed how AI serves as an "Iron Man suit" for navigating complex biological data:

• Bioinformatics Analysis: Using AI to parse CSV files containing gene expression counts from bulk RNA sequencing.
• Agentic Workflows: Deploying AI agents to perform 30-minute literature reviews, formulate hypotheses, and write Python code to analyze patient-specific blood data (e.g., monitoring for CHIP—Clonal Hematopoiesis of Indeterminate Potential).
• Target Discovery: Identifying unique protein targets like PENX3, which was expressed 10,000 times more in the tumor than in healthy tissue. AI helped identify this target, which had been missed by traditional research because it is hydrophobic and not easily detected in standard water-based assays.

3. Personalized Treatment Vignettes

• mRNA Cancer Vaccine: Encoded with specific mutations from Sid’s tumor to prime his immune system. Project timeline from start to injection was only six months.
• TCR-T and CAR-T Therapy: Engineering T-cells to recognize specific cancer markers.
• Logic-Gated CAR-T: To prevent off-target toxicity (e.g., liver damage), they utilized an "AND-gate" design. The therapy only activates if two specific proteins (B7H3 and FAP) are present, ensuring the "nuclear bomb" of the treatment does not destroy healthy organs.

4. Key Arguments and Perspectives

• Incentive Misalignment: Sid argues that doctors are incentivized to minimize liability, whereas patients must prioritize survival. He advocates for patients to be proactive, well-informed advocates who challenge the "standard of care" when it is insufficient.
• Democratization of Knowledge: AI allows non-specialists to become competent enough to engage with experts, effectively lowering the barrier to entry for complex medical decision-making.
• Clinical Trial Reform: The speakers argue that the current $1B+ cost to bring a drug to market is unsustainable. They propose:
  • Adopting notification-based trial approvals (similar to Australia).
  • Granting researchers "IRB freedom" to choose independent, efficient ethical review boards rather than being forced to use hospital-specific boards.

5. Notable Quotes

• "I rather die from a treatment than from the cancer. Dying from cancer is a really miserable way to go." — Sid Severy
• "AI is amazing at helping you suggest things to discuss with your oncologist... it gives me a rapid way to get up to speed and start to understand the sort of circumstances around this disease." — Jacob Stern
• "Curing cancer may happen one patient at a time." — Chris Nicholson (synthesizing the speakers' approach)

6. Synthesis and Conclusion

The core takeaway is that the future of medicine lies in personalized, data-driven, and AI-accelerated interventions. By treating a medical crisis like a startup—collecting maximum data, applying first-principles logic, and using AI to bridge knowledge gaps—patients can significantly improve their outcomes. The speakers are now working to "pave the road" for others by launching companies (e.g., Veas, Ardan) that aim to standardize these diagnostic and treatment workflows, moving them from experimental "Roadster" models to accessible, scalable solutions for the broader population.