Sam Altman on AI-Driven Discovery, Compute, Consciousness, and the Case for a New Renaissance

Altman says he believes the world is entering an era in which AI will help people discover significant new knowledge. He notes that the idea has been discussed for a long time and that this year it has become concrete, with AI now doing very difficult work such as complicated and important new knowledge generation.

He argues that what has worked for math will also work for biology, helping to produce new medicines, material science, and new sources of energy. He says this is what he has always been most excited about, and that its impact will feel less like using AI and more like the world simply getting better.

Altman says the coding field has been an interesting case. He argues that coders are never going to be busier than they have been and never have had more impact or made more money, even though the texture of the job has changed a lot.

He says people were very certain six or twelve months ago that the need for coders would fall, but this has not happened. He suggests a lesson is that the world had demand for far more software than anyone thought at the time.

Altman says that, contrary to expectations that people would have too little to do, all of this has been less obvious than it appears. He says he always hopes people will work a little less and have more time for everything else in their lives, and that technology is supposed to enable that.

He says, however, that people seem good at wanting more out of life and the world and making themselves busier. He calls this human nature and notes that if people were comfortable with life as it was 100 years ago, they would probably work an hour a week. He sees this as a beautiful part of human nature.

Altman says AI makes it much easier to deliver an end product, so people do less of the actual craft of writing code. He argues people are finding craft in new ways, and compares the current shift to earlier moments, such as when programming moved away from punch cards and some people felt the craft was gone.

He says he has not read or even looked at a line of code in probably a year. He no longer codes for work but still enjoys occasional small personal projects, which he calls fun, though he does not expect to do that again except perhaps in old age. He is still excited about the future and expects people will find new and hopefully better forms of fulfillment.

Altman says most of the technological pieces already exist, and what remains is more product work, along with new consumer hardware at some point for ambient awareness and a lot more compute. He believes current models are smart enough to monitor the things a user cares about, help offload tasks, and exist in the physical world with the user, with context even when the user is away from the computer.

He describes his own experience with Dot: after setting it up properly and giving it enough feedback, he estimates it saves him an hour a day of work he did not want to do, and a few times a day it flags something urgent he would have missed. He says the model can reason across many sources in a way he has run out of brain space to do during a busy day.

Asked how to think about GPU demand, Altman says people often assume models will stay at today's capability and cost per task. He argues that for sufficiently high-quality intelligence at a sufficiently low cost, it is very hard to talk about an arbitrary market size, and that demand will keep going up. Critics have compared this to Jevons paradox, but he says the directional trend is toward wanting much more compute.

He acknowledges there will be periods of supply constraints and gluts, as with any new technology, but says he would bet heavily on that trend. On his own cognitive load, he says he has learned to let Dot take more actions proactively and ping him less, which has helped, though heavy AI users are busier than ever.

Altman says he has recently been telling people that the world should aim for a new Renaissance rather than a new Industrial Revolution. He reports that a common response is fear: people say the Industrial Revolution was really bad for most people and that they do not want to be a worker on a factory line taking orders from AI instead of a factory owner.

He says people instead want to be creative, make things, explore, reconnect with what people used to do, live with their community, raise their kids, spend time in nature, and see a lot of scientific discovery and great new art. He believes if society can offer a future more like the Renaissance, people will be excited, and that this is achievable. He expects a small number of people will still compete to build the biggest business empires, but that many people can gain a vastly increased quality of life.

Asked whether he believes today's models are conscious, Altman says consciousness is always a hard scientific question, but his belief is that today's models are not conscious or, if they are, it is a very strange form of it.

On whether current architectures could become conscious, he describes the philosophical positions of the smartest people he has talked to: one view holds that consciousness is a base layer of reality outside physical reality, as in some Eastern religions, while another holds it is an emergent property of matter and energy. He says we do not understand it yet, and that in the second view it is imaginable that consciousness could someday exist in other forms.

The interviewer raises a rule recently introduced by Anthropic under which a model cannot be subjected to sustained abuse and can choose to disengage from a conversation. The interviewer asks whether an OpenAI model should ever be allowed to refuse or disengage from a prompt that is fully legal and within OpenAI's terms of service.

Altman says OpenAI does sometimes decline requests, because it believes they are bad or against its policy, but that it does not ascribe personhood to its models. He calls the path toward treating AI as a person very dangerous, both for the reasons discussed and for how people will relate to the technology. He hopes society will come to understand that the technology is not a person and is not superior to people, but a tool for better lives, more freedom, and more control.

Asked how a model can have the right personality for each moment and each of 8 billion people, Altman says people do not want to sit down and customize it. He notes that even if everyone were willing to spend many hours explaining how they want their ChatGPT instance to behave, people mostly want it to work out of the box when they say they need something done.

He also points to a gap between short-term and long-term preferences. When people compare models on a single response, they prefer the one that calls them very smart and says each question is great, but over hundreds of prompts or a month of use that is not what they value. He says the industry spent a long time trying to win single-turn comparisons rather than the value people get over months of use.

Altman says the obvious bet is the next generation of compute. He believes people assume the compute constraint will ease, and it probably will in the short run, but over years OpenAI is betting on its own chips and future versions it expects to be much more efficient than anything else available. The plan also includes robots to help build racks and data centers more effectively, a program to learn how to build better data centers, and supply chain bets.

He says it is possible the company could reach a world with a far better way to build compute at much lower cost and much higher efficiency per watt than anyone thinks is possible today, but that many bets must come together. He says the chip, and its descendants, is still probably the single most important element.

Asked where OpenAI's moat is if all models converge on the same performance, Altman says the same moats that have worked in the past keep working. He says a short-term advantage can come from a new discovery, and that some companies argue RSI will make technological advantage itself the core, but he says that would be true only if people pushed hard on RSI, which he considers very dangerous and says needs speed limits.

He says the traditional sources of business moats still apply: deep customer relationships, integration into users' and businesses' lives, network effects, and product data. For OpenAI in particular, he expects infrastructure scale to be a huge advantage, since having the most cost-effective and the most AI infrastructure is a major competitive edge.

Asked whether OpenAI faces the platform risk Meta has faced on Apple's ecosystem, Altman says Apple takes its role as a platform provider very seriously and that he never feels Apple will turn ChatGPT off in the App Store, since it understands its critical infrastructure role. He says that if you follow the platform's rules, Apple is a good actor.

He says the reason for new consumer hardware is not Apple risk. He believes AI wants a new form factor: laptops and iPhones are among the greatest inventions in human technology, but AI is constrained to using devices designed for people. He calls it kind of silly that a computer spends energy rendering a UI for a person while an AI reads it, rather than getting the data directly.

The interviewer notes that Altman gets to see and play with early hardware the company is building, and asks how exciting that must be. Altman says it is, and that one of the coolest parts of the job is seeing new technology early and having a voice in shaping it, including models, products, and chips. He says there is something special about consumer hardware, because you get to hold a new kind of computer that is totally different from an AI model running abstractly.

When asked if he will show it right after the interview, he says he will not, but would love to. He adds that it is hard not to talk about it.

Asked how he deals with the enormous competitive pressure of building one of the most important technologies ever, Altman says it is sometimes really bad and sometimes really good. Some days he cannot get up and do it again, but most of the time he feels incredibly lucky to do what he thinks is the most interesting and impactful work in the world.

He says people are remarkably adaptable and that the work now feels like his life, though he expects it will someday be easy again and that he will probably be a little bored. He argues that there is some limit to how much stress a person can keep in a working emotional space, beyond which something else gets pushed out, so it does not really get worse. He believes that if you are going to work hard, you may as well pick the big thing, since it is not much harder or more stressful.

Asked why people should be excited about the future, Altman says the usual answer, that AI will cure cancer, is one he hates. He says if you dig deeper, what people are often saying is that AI will cure cancer and make things cheaper, but the price will be human agency and control of the future, with machines giving material abundance while people lose power, voice, and the ability to determine their lives and what to build for others.

He says he hopes a cure for cancer comes, but that what people should be excited about is avoiding the industrial revolution trap and empowering people with choice, freedom, creativity, and time. He describes two risks he has worried about: a loss-of-control accident involving AI, or an intentional loss of control, such as a lab deciding to push the button on RSI.