Sam Altman thinks AI will get “new ideas” next year

On Tuesday, OpenAI CEO Sam Altman published a new essay titled The Gentle Singularity, in which he shares his views on how artificial intelligence will change the human experience over the next 15 years.

The essay continues Altman’s tradition of futuristic musings: he praises the potential of AI up to the level of Artificial General Intelligence (AGI), while arguing that OpenAI is already close to achieving that goal, but at the same time softening the timing of its arrival. Altman makes such publications regularly, describing a future in which AGI transforms labor, energy, and the social contract. It is not unusual to find hints of the company’s future developments in these texts.

At one point, Altman mentions that as early as 2026, the world is likely to see AI systems “capable of finding fundamentally new ideas.” While the wording remains vague, OpenAI representatives have previously stated that they are focused on teaching AI to generate original, useful ideas about the world.

At one point, Altman mentions that the world is likely to see AI systems “capable of generating original, useful ideas about the world.”

At the announcement of the o3 and o4-mini models in April, OpenAI co-founder and president Greg Brockman said the models were the first time scientists had used them to generate new and valuable ideas.

According to Altman’s blog, OpenAI will soon increase its focus on creating AI capable of scientific novelty. But the company is not alone: its competitors are also looking to create AI that can formulate hypotheses and contribute to new scientific discoveries.

At the same time, OpenAI’s competitors are also looking to create AI that can formulate hypotheses and contribute to new scientific discoveries.

In May, Google published a paper about AlphaEvolve, an AI programming agent that is said to have developed new approaches to complex math problems. FutureHouse, a startup backed by former Google CEO Eric Schmidt, claimed its AI had made a genuine scientific discovery. Anthropic, for its part, has launched a program to support scientific research.

If the effort succeeds, AI could automate a key part of the scientific process and enter industries such as drug development, materials science and other science-oriented fields.

This isn’t the first time Altman has hinted at OpenAI’s plans via his blog. In January, he wrote that 2025 would be “the year of agents” – shortly afterward, the company unveiled its first three AI agents: Operator, Deep Research, and Codex.

The company’s first three AI agents are Operator, Deep Research, and Codex.

Nevertheless, creating AI capable of genuinely new discoveries may prove much more difficult than developing agent-based systems. There remains skepticism in the scientific community about AI’s ability to generate truly original ideas.

Skepticism remains in the scientific community about AI’s ability to generate truly original ideas.

Earlier this year, Hugging Face’s chief scientist Thomas Wolfe wrote an essay arguing that today’s AIs don’t know how to ask great questions – which is the key to any scientific breakthrough. Former OpenAI researcher Kenneth Stanley also noted that current models are incapable of generating qualitative hypotheses.

And that current models are incapable of generating qualitative hypotheses.

Now Stanley is working on building a team at startup Lila Sciences, which has received a $200 million investment. The company aims to create an AI lab focused on generating scientific hypotheses. He says this is a particularly challenging task because it requires AI models to understand what might be considered creative and interesting.

And he says it’s a particularly difficult task because it requires AI models to understand what might be considered creative and interesting.

Whether OpenAI will actually succeed in creating an AI capable of original scientific insights remains to be seen. But Altman’s new essay may once again provide a hint of where the company is headed.

The Sam Altman thinks AI will get “new insights” next year was first published on ITZine.ru.