Fireside Chat, Interview
Sam Altman: What Startups Will be Steamrolled by OpenAI? Where is Opportunity?
- OpenAI anticipates a steep trajectory of improvement where current model shortcomings will be resolved by future generations, enabling rapid progress in the O series, multimodal capabilities, and image-based models that scale with new inference time paradigms.
- The organization plans to shift focus from individual models to systems over the coming year, with O1 demonstrating agentic capabilities capable of executing multi-day or multi-week tasks with minimal supervision, eventually functioning as smart senior co-workers.
- No-code tools are expected to evolve in two stages: initial tools enhancing coder productivity followed by high-quality platforms for building full startups, though the latter remains several years away.
- Strategic planning encompasses five-year and ten-year horizons, projecting an unbelievably rapid rate of technological improvement driven by scientific progress, while expecting societal changes to outperform expectations despite appearing minimal in the near term.
- Significant economic impact is predicted, including trillions of dollars in new market cap and value creation in healthcare and education, driven by products previously impossible or impractical to build.
- OpenAI intends to price based on compute consumption for user problems rather than per seat or per agent, leveraging a user base of hundreds of millions to amortize high training costs while ensuring revenue justifies investment.
- Market entrants face a complex, expensive industrial process distinct from the internet revolution, with foundation model entry costs estimated under $100 billion; however, businesses built solely to patch model shortcomings risk obsolescence as models improve.
- Risks include the complexity of the entire ecosystem, uncertainty in specific product strategy details, and top-10% concerns regarding semiconductor supply chains and international tensions, requiring careful balancing of power, networking, and chip delivery.
- OpenAI plans to maintain a strong open source ecosystem alongside well-integrated services and APIs, with scaling laws for capability improvement expected to hold true for a long time.
- Positive compounding effects in training are anticipated as teams gain proficiency, while the market has internalized the rate of model improvement, reversing previous trends of betting against progress.