In a new test published Tuesday entitled “Sweet singularity,The CEO of Openai, Sam Altman, shared his latest vision for the way IA will change human experience over the next 15 years.
The test is a classic example of Altman’s futurism: setting up the promise of AGA – and arguing that his business is quite close to the feat – while simultaneously minimizing his arrival. The CEO of Openai frequently publishes tests of this nature, properly presenting a future in which AGS disturbs our modern conception of work, energy and social contract. But often, Altman’s tests contain advice on what Openai then works.
At a time of the test, Altman said that next year, in 2026, the world “will probably see the arrival of [AI] Systems that can find new ideas. Although it is somewhat vague, Openai leaders have recently indicated that the company focuses on AI models to find new interesting ideas on the world.
During the announcement O3 and O4-Mini of Openai reasoning models In April, the co-founder and President Greg Brockman said it was the first models that scientists had used to generate useful new ideas.
Altman’s blog post suggests that in the coming year, Openai himself can increase his efforts to develop an AI that can generate new information. Openai would certainly not be the only company focused on this effort – several of OpenAi competitors have moved their attention to the training of AI models which can help scientists offer new hypotheses, and therefore new discoveries on the world.
In May, Google published a Paper on alphaevolveAn AI coding agent that the company claims to have generated new approaches to complex mathematical problems. Another startup supported by the former CEO of Google, Eric Schmidt, Futurehouse, says that his AI agent tool has been capable of making a real scientific discovery. In May, Anthropic launched a program to support scientific research.
In case of success, these companies could automate a key element of the scientific process and potentially enter massive industries such as the discovery of drugs, the science of materials and other fields with the basis.
It would not be the first time that Altman would have tipped his hat on Openai’s plans in a blog. In January, Altman wrote Another blog article suggesting that 2025 would be the year of agents. His company then filed its first three AI agents: the operator, deep research and codex.
But bringing AI systems to generate new ideas can be more difficult than making them agents. The wider scientific community remains somewhat skeptical of the AI’s ability to generate truly original ideas.
Earlier this year, Thomas Wolf, Director of Sciences of Hugging Face, wrote a test arguing that modern AI systems cannot ask big questions, which is the key to a great scientific breakthrough. Kenneth Stanley, a former OPENAI research official, also told Techcrunch that today’s AI models could not generate new hypotheses.
Stanley is now a team from Lila Sciences, a startup that has collected $ 200 million to create a laboratory powered by AI specifically focused on AI models to find better assumptions. This is a difficult problem, according to Stanley, because it involves giving meaning to AI models for what is creative and interesting.
It remains to be seen if Openai creates an AI model capable of producing new information. However, Altman’s test can have something familiar – an overview of the place where Openai is probably heading afterwards.
