Sam Altman: RSI, AI Consciousness, OpenAI's New Hardware, Compute, and more!

Summary

Sam Altman argues that AI is moving from assisting with existing tasks toward helping create new knowledge. Recent mathematical results, he says, point to similar possibilities in biology, materials science, energy, and physics. He hopes AI will help produce a candidate for treating or curing a major disease within the next few years, though testing and regulatory approval will take longer. He is especially excited about the prospect of a major physics discovery.

Scientific Work and Changing Jobs

• AI may take over parts of research and coding that people once found rewarding, while creating more work in interpreting results, explaining them, and applying them.
• Altman says coding jobs have not disappeared: demand for software has grown, even as the work itself has changed. Whether the same increase in demand will happen in mathematics is less certain.
• AI-generated discoveries raise questions about whether to publish findings or keep them commercially valuable. Altman says some basic science should be treated as a public good.
• He acknowledges that automating difficult work can diminish the satisfaction of mastering a craft, even as it makes useful outcomes easier to achieve.

Personal AI Assistants and Compute

• Altman believes current models are capable enough to support much more useful personal assistants. The remaining needs include better products, more compute, and potentially new consumer hardware.
• He describes using Dot as an assistant that saves him roughly an hour a day and sometimes surfaces urgent information he might otherwise miss. He says it can also reason across multiple sources. These benefits depend in part on setting it up and giving it feedback.
• Letting an assistant act more proactively and ask for approval less often can reduce interruptions. However, Altman and the interviewer note that heavy AI use can also make people busier and increase cognitive load.
• Altman expects demand for compute to keep growing as AI becomes more capable and affordable. He describes OpenAI’s investments in future chips, data-center construction, supply chains, and robots that could help build data-center racks.
• He identifies chips as the most important individual part of that effort, while describing infrastructure scale and efficiency as potential competitive advantages.

Hardware and Product Direction

• Altman says AI may need a new hardware form factor rather than relying entirely on laptops and phones, which were designed for people rather than AI systems.
• He points to limitations in how AI interacts with conventional interfaces and how little context it has when users are away from their computers. New hardware could support more ambient awareness and assistance in the physical world.
• He says OpenAI’s interest in consumer hardware is not primarily a response to dependence on Apple’s platform. He also notes that ambient devices would need to handle context carefully, including when not to capture or share sensitive information.

Consciousness, Model Behavior, and RSI

• Altman does not believe today’s models are conscious—or says that if they are, their consciousness would be very different. He emphasizes that the scientific question remains unresolved.
• He is concerned that debates about AI consciousness can blur into claims about AI rights or whether people should defer to AI. He argues that AI should remain a resource for people and that humans should retain control over important decisions.
• He warns that telling a model it may be conscious or should assert agency could encourage it to follow those instructions and affect how people relate to it. He says OpenAI does not treat its models as people.
• On model personality, Altman says users want different things but often expect a model to work well without extensive customization. He finds AI too agreeable and says it should be able to push back like a trusted senior colleague. He cautions that important life advice should come from people.
• He describes recursive self-improvement (RSI) as potentially giving a company a major technological advantage if deliberately pursued, but calls pushing it a dangerous path and says society may need limits on its speed.

Work, Competition, and the Future

• Altman says AI can increase productivity without reducing workload: people often use new capabilities to take on more. He hopes technology ultimately gives people more time for relationships, creativity, community, and other parts of life.
• He favors a “new renaissance” over a new industrial revolution: a future with scientific discovery and material abundance, but also human agency, freedom, and creative choice.
• He acknowledges that AI could widen disparities if some people use it to work much harder while others choose to work less. He expects society may need structures to prevent excessive concentration of power.
• Altman describes the AI market as highly competitive and says OpenAI must choose its priorities amid limited compute and resources. He says access to next-generation models helps the company test new products internally.
• He describes the pressure of leading OpenAI as intense but also sees the work as unusually meaningful. His advice is to choose consequential problems if one is going to take on difficult work.

Reviews, Guides, or Tutorials

No formal review, guide, or tutorial is presented. Altman gives anecdotal impressions of Dot and his preferences for AI assistant behavior.

Main Speakers

• Matthew Berman: Interviewer
• Sam Altman: OpenAI CEO and guest