You don't need to write prompts for ChatGPT, Sam Altman reveals a secret

Sam Altman told students that stronger AI systems have changed the scale of what one person can build. He said the biggest openings may lie in problems made newly possible by automated coding and scale.

In Short

• Altman said scale often creates unexpected value beyond early assumptions
• At Y Combinator, larger startup batches gained strength from internal networks
• OpenAI found GPT-3 users were informally turning API access into chats

For students learning computer science, the question is no longer simply what problem they should be asked to solve. In a recent discussion, OpenAI CEO Sam Altman argued that the arrival of increasingly capable AI systems has changed the scale of what one person can attempt - and, with it, the way students ought to think about problems.

“The level of ambition you can have, the speed at which you can move, the amount of stuff you can do at once, it’s just totally different,” Altman said.

He questioned the value of assigning students a specific problem to attack. If a problem or startup idea is obvious enough for a teacher to identify, he said, it may already be obvious to many others. The more interesting opportunities, in his view, may be those that were not possible before automated coding and AI systems became available.

“When we started OpenAI, we were, generously speaking, one of maybe four AI efforts in the world,” Altman said. He suggested that there may now be opportunities that were impossible before the automated coding era, and which only a small number of companies are pursuing.

But he acknowledged that he does not know what those opportunities are.

“It’s much more likely you all know what that is than I know what that is,” he told the students.

SCALE CHANGES THE PROBLEM

Altman returned repeatedly to the question of scale, describing it as one of the most persistent lessons of his career.

He said the most interesting developments he had observed often involved “emergent properties that scale” or situations in which scaling continued to produce returns well beyond what people had expected.

He recalled his experience at Y Combinator, where there was once an argument that the programme had become too large and should fund fewer companies. The assumption was that the strongest startups would be obvious and that expanding the batch would add little.

But, according to Altman, something else emerged: the network within the larger group itself became valuable.

“That was an emergent property at scale that just hadn’t been discovered before,” he said.

The same lesson, he argued, appeared in AI. When OpenAI began scaling models, the obstacles were not merely technical. There were questions about computing power, capital, engineering resources and whether the investment could be justified.

“Stuff breaks at an accelerating rate and in an unpredictable way as you scale it,” Altman said.

For that reason, scaling requires more than ambition. Each difficulty has to be separated and addressed, he said, while the organisation remains committed to a clear objective.

WHAT CHATGPT TAUGHT OPENAI

Altman also described how OpenAI arrived at ChatGPT, saying the company initially struggled to find a product around GPT-3.

The model was made available through an API in 2020, but initially attracted little attention. Developers later began using it in unexpected ways, including as a conversational system.

“We had been thinking, thinking, we just couldn’t do it,” Altman said of finding a product for GPT-3.

The company noticed that although developers were struggling to build businesses around the API, many were using their API keys simply to chat with the model. OpenAI then built a chatbot around that behaviour.

It was intended, Altman said, largely as a research demonstration. Instead, it went viral.

“We have the potential here at a killer product,” he recalled thinking after watching its growth.

That experience became an example of another principle Altman had learned at Y Combinator: watch what users are doing, particularly when they begin using a product in ways its creators did not anticipate.

AI AS A NEW UTILITY

Altman sees a broader change beyond individual products. He compared the possible development of AI to the emergence of electricity and the internet as utilities.

The difficulty, he said, is explaining AI to people when its uses are still being discovered. Electricity companies did not initially sell consumers the idea of “electricity”; they sold what electricity enabled - such as light at night.

He believes AI may follow a similar path.

“I think what is happening is we are in the process of creating a new utility,” Altman said.

For consumers and businesses, he expects the underlying hardware to become less important than access to the intelligence itself — much as people pay for mobile or internet services without thinking about the particular hardware carrying the connection.

And for students deciding what to build, his suggestion was pointed: rather than waiting for someone to assign the right problem, look for what becomes possible when the technology changes the limits of what one person can do.

- Ends