AI Summary
→ WHAT IT COVERS Sarah Guo, founder of Conviction, discusses the beliefs and concerns of approximately 250 frontier AI researchers and entrepreneurs, covering compute independence, open source model policy, investment decision-making frameworks, and why regulatory barriers—not technical capability—represent the primary obstacle to US AI competitiveness. → KEY INSIGHTS - **Technology-forward investing:** Rather than starting from customer problems alone, Conviction evaluates which workflows and professions structurally match current model capabilities. This led to early investment in Harvey, reasoning that law is fundamentally structured language requiring document retrieval and text generation—a direct fit for 2022-era models—before the thesis became consensus. - **Researcher disempowerment at scale:** A notable shift within the past 12 months shows frontier lab researchers increasingly believing either that their individual contributions are irrelevant or that compute scale alone determines outcomes. Both conclusions reduce personal ownership of results, creating a psychological dynamic that affects retention and motivation at the largest AI labs. - **Compute independence as strategic priority:** Working backward from a functioning data center reveals a global supply chain with critical single-point vulnerabilities—TSMC controls key components, energy infrastructure lags demand, and nothing moves the needle for hyperscalers before 2030. Conviction is actively investing in nuclear energy, alternative chip architectures, robotics, and data center labor solutions. - **Investment conviction process:** Guo starts at an 8-9 conviction rating instinctively on people, then works backward to identify gaps. She writes full investment memos early, solicits external second reads from trusted investors, and explicitly identifies what evidence would change her position—treating the process as hypothesis falsification rather than consensus building. - **Open source model policy:** Restricting open source AI models in the US would only constrain law-abiding American businesses while leaving adversarial actors unaffected. The productive alternative is rigorous safety testing—including systematic research into backdoor behaviors in Chinese models—combined with accepting that broad, cheap intelligence access is economically necessary for US industrial competitiveness. → NOTABLE MOMENT Guo describes passing on Suno, the AI music generation company, despite knowing the founder through a mutual contact. She underestimated how many people want to create music, calling her own intuition wrong—a candid admission that even domain-focused investors misread consumer behavior in emerging AI categories. 💼 SPONSORS [{"name": "Ramp", "url": "https://ramp.com/invest"}, {"name": "Rogo (Felix)", "url": "https://rogo.ai/felix"}, {"name": "WorkOS", "url": "https://workos.com"}, {"name": "Vanta", "url": "https://vanta.com/invest"}, {"name": "Ridgeline", "url": "https://ridgeline.ai"}] 🏷️ AI Investment Strategy, Compute Independence, Open Source AI Policy, Frontier AI Research, Early Stage Venture Capital
