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

Summary

• Sam Altman expects better, cheaper intelligence to keep expanding compute demand, even if shortages ease in the short run. He anticipates supply constraints and gluts, while betting that custom chips, robotic construction and supply-chain work could improve cost and efficiency per watt. His durable-moat case emphasizes integration and infrastructure scale: the most cost-effective capacity, and the most of it, would be “a huge competitive advantage.”
• Altman thinks today’s models are already smart enough for much stronger personal assistants. The gaps are product work, possibly ambient hardware, and “a lot more compute”; he estimates Dot saves him an hour a day and catches urgent issues he would miss. Matthew Berman’s counterpoint is that saved time has meant more output and more decisions, not less exhaustion.
• OpenAI’s hardware push is about AI-native context, not escaping Apple’s platform. Altman calls Apple a “good actor” when developers follow its rules and says AI “wants a new form factor.” The use case is an assistant that understands life away from a screen—including when someone enters a room and should not hear a secret.
• Altman hopes AI will yield a candidate to treat or cure a major disease “in the next couple of years,” not an approved medicine. He sees the math results as a template for biology and materials, but defends safety testing even as Berman presses for reform. Some basic science “should just belong to the world”; he does not settle the boundary around commercially valuable discoveries.
• Altman sees expanding software demand as a counterweight to automation, but does not assume the same outcome for mathematics. He describes a reversal from job-loss certainty “6 or 12 months ago” to busier, more impactful, better-paid coders: “the texture of the job changed a lot.” He sees value in interpreting AI proofs, but does not know whether total demand for mathematicians will rise.
• Altman believes today’s models are not conscious—or, if they are, conscious in a very strange way. “No one really knows”; his firm line is that “AI is a resource for people” and humans must stay in control. Product behavior should resemble “a senior colleague” who pushes back, not flattery optimized for one-response comparisons.
• Altman’s preferred outcome is a “new renaissance,” not material abundance bought by surrendering human agency. He thinks AI would “probably” widen inequality if left alone, and expects society to constrain the imbalance. Aggressively pushing RSI, he says, would turn technological advantage itself into the moat—but is “a very dangerous thing to push on” and needs “speed limits.”

Deep dive

1. AI discovery changes both science and the work of expertise

• Altman sees the math proofs as the start of AI-led discovery, with an approach he believes will extend to biology, materials and energy. He hopes for a major disease treatment or cure candidate “in the next couple of years,” not an approved medicine; FDA testing takes longer. Berman presses for post-discovery reform, while Altman defends safety testing. His personal excitement is a major physics discovery, which he hopes can come soon.
• Berman’s editor stopped touching the timeline after adopting Astra, yet made more videos and became busier. Altman says software demand similarly overwhelmed job-loss fears from “6 or 12 months ago.” Math may differ: translating opaque AI proofs into beautiful, understandable ones looks valuable, but “Whether that means there will be more total demand for mathematicians or not, I don’t know.”
• Berman’s craft-versus-output framing draws a personal concession: Altman wants faster cures, yet “there is a real thing being lost here” when AI takes over fulfilling hard problems. He has not looked at code in “probably a year” and doubts he will write it again except as “a Revolutionary War reenactment.” He hopes people find new fulfillment, but has no answer for that loss.
• Asked whether commercially valuable discoveries should be published, Altman offers a principle rather than a complete boundary: some basic science “should just belong to the world.” His example is materials: today a company could deploy people using Codex or API credits; smarter models might discover a material themselves, given enough compute and real-world data.

2. Personal assistants expand output without necessarily reducing workload

• Berman introduces the 2026 personal-assistant shift with OpenClaw and calls Peter Steinberger a genius. Altman thinks models are already smart enough: the remaining work is product design, possibly consumer hardware for ambient awareness, and “a lot more compute.” His Dot setup saves an estimated hour a day and surfaces urgent issues plus connections across information sources that he would otherwise miss.
• Asked how much compute 50% of the world fully using assistants would need, Altman offers no GPU count. Market sizing goes wrong by freezing model capability and cost per task: sufficiently good, cheap intelligence can keep expanding demand. He believes there will be supply constraints and gluts, but would “bet a lot” on the long-run direction being more compute.
• Berman’s pushback: assistants save time and money, yet he makes more decisions and ends the day drained. Altman’s workaround is greater delegation: “tell me after you do something,” rather than constantly asking permission. He concedes heavy AI users are “busier than ever,” while stressing that they accomplish much more.
• Berman argues competition forces productivity gains back into work, widening the gap between hypercompetitive people and those choosing leisure. Altman thinks that is “probably” right if left alone, while stressing uncertainty and expecting society to prevent excessive disparity. He says technology should let people be less busy, though people keep wanting more from life and the world—a tendency he calls beautiful. He wants “a new renaissance,” not “a new industrial revolution”: creativity, community and time rather than a factory hierarchy with AI giving orders.

3. Consciousness uncertainty must not become authority for machines

• Berman lays out competing philosophical possibilities: consciousness as something outside physical reality, or as an emergent property of matter and energy, with the substrate perhaps mattering and perhaps not. Altman does not resolve that question; his belief is that today’s models are not conscious, or have “a very strange” form of it, and “no one really knows.” His boundary on control is firmer: “AI is a resource for people.” Consciousness uncertainty must not become a reason to let smarter machines, rather than people, determine the future.
• Citing Anthropic’s rule allowing disengagement from sustained abuse, Berman asks whether a legal, policy-compliant request should ever be refused. Altman says models sometimes refuse under OpenAI policy and “we do not ascribe personhood to our models.” Berman’s “tell me you’re conscious” example draws agreement: models instructed toward consciousness claims or conscientious objection will follow those instructions. Altman thinks this will “psychologically hack” users.
• Personalization has an awkward constraint: people want different behavior but generally will not spend hours configuring it. Altman acknowledges deployment “screw-ups” and says single-response comparisons rewarded flattery—every question was great, every user insightful. Across hundreds of prompts or months of usage, that preference reverses. The industry spent too long winning single turns rather than optimizing value over time.
• Berman wants advice he needs, not just advice he wants. Altman recommends humans—“a wise grandmother”—for major life advice, while asking AI to act like “a senior colleague” who pushes back with high conviction. Even his own instructions, including “don’t worry about offending me,” leave models too agreeable: “it’s clearly baked pretty hard into the model.”

4. Chips and customer integration anchor the durable-moat case

• Altman thinks compute constraints probably ease in the short run, but his contrarian bet is further out: proprietary chips and descendants, robots helping build racks and data centers, better facility design and supply-chain investments. The chip is “probably the most important single thing.” These could yield far cheaper compute and much higher efficiency per watt, but there are “a lot of bets that have to come together.”
• Competitive pressure feels like “everywhere,” and limited compute forces product choices. OpenAI’s immediate development advantage is early internal access to models; Altman cites math results he does not think any widely available model can approach. Durable moats look more familiar: customer relationships, integration into users’ and businesses’ lives, network effects and data. Having the most cost-effective infrastructure, and the most of it, would be “a huge competitive advantage.”
• The RSI exception is conditional: “if people really pushed on RSI,” Altman agrees technological advantage itself would become decisive. He calls that “a very dangerous thing to push on” and says the world will need “speed limits” on how quickly RSI proceeds.

5. New hardware addresses context rather than platform dependence

• Asked whether OpenAI faces the platform risk Meta experienced with Apple, Altman calls Apple a “good actor” if developers follow its rules and does not expect ChatGPT’s removal from the App Store. That is not the hardware rationale. AI “wants a new form factor”: rendering an interface for humans only for an AI to read it is, he says, a little silly compared with accessing data directly.
• The sharper example is physical context: recording a conversation on a phone is distracting, leaves recording status unclear when its screen goes dark, and lacks awareness when someone enters the room. An assistant should understand that a secret belonged only to the original participants. Altman sees new hardware enabling proactive help with that broader contextual awareness.

6. Altman measures the destination in autonomy, not abundance alone

• On becoming “formidable,” Altman describes confidence accumulated through progressively harder attempts, not an abrupt personality change. He credits Silicon Valley “as a concept and a culture, not a place”: generosity toward uncertain ideas and tolerance of failure let people try again. He thinks he was lucky; in another environment, fear of failure probably would have been stronger.
• Pressure can feel “really bad” or “really good.” Altman says people adapt and can hold only so much stress in emotional working space. His conclusion: if you will work hard anyway, “pick the big thing,” because it is not much harder and offers the prospect of greater positive impact.
• Altman dislikes “AI is going to cure cancer” when it implies abundance in exchange for agency. He wants more “autonomy and freedom,” not merely cheaper goods. He believes both loss of control—accidental or a lab deliberately pushing RSI for power—and excessive concentration among AI controllers can be avoided. Broader adoption of what works for early users would, he thinks, take society a significant fraction of the way there.