In a fresh, thought-provoking article entitled “The Gentle Singularity,” Openai CEO Sam Altman moved in with bold predictions once again. By 2026, artificial intelligence systems can create “new insights.” This claim may sound vague in itself, but in the context of Openai's development goals and the broader competitive scene of AI research, it is heavy. Famous for his futuristic vision and bold declarations about artificial general information (AGI), Altman presents this prediction as part of the expanded evolution of human society where AI remakes work, energy, innovation and even scientific breakthroughs.His essay, released on June 10, 2025, is more than predicted. It can be used as an unofficial guide to where Openai and its competitors move on. This race towards AI systems that not only assemble current information but come up with real new ideas is becoming the next arms race for technological innovation. With major breakthroughs being achieved by startups like Google Deepmind, Anthropic, and startups like Lila Sciences and Futurehouse, interests are high not only for the hegemony of the company, but also for the future of science itself.
What Sam Altman really said: AI with “new insights” by 2026
Altman's essay describes his vision for “serene singularity,” an era in which AGIs become productive and evolutionary partners rather than destroying or dismantling civilization. Perhaps one of the most notable things he predicts is this:“By 2026 we will see the arrival of a system that will allow us to understand new insights.”This effectively problematic statement has great meaning. First of all, Openai shows that artificial intelligence is approaching the level of modeling ability that allows you to move past summary, prediction, or pattern matching and start thinking creatively. This is simply the distinction between AI that reports data and AI that meaningfully adds to human knowledge.Openai President and co-founder Greg Brockman recently supported this concept when he launched the O3 and O4-Mini models in April 2025. “These were the first models scientists were used to create useful new ideas in their field,” he declared, suggesting that Openai was already developing the kind of AI that Altman described in his essay.
What does “new insight” mean?
“Final Insights” in AI are the capabilities of AI models, the ability to generate ideas and hypotheses that evoke unprecedented ideas. This feature extends well beyond the current realm of AI chatbots and large language models' capabilities, which are based on spitting known facts.
AI Race for Survival: Other Technology Titans Are Not Lagging
Altman's prophetic essay is not isolated. Over the past few months, many key players have taken strategic action in line with the premise of creating “new insights.”
- Google Deepmind has released a paper on Alphaevolve, a coding AI agent that has created innovative solutions to challenging mathematics problems.
- The founders of Claude AI's humanity established a research grant program in May 2025 to support research projects that employ AI to generate scientific hypotheses and carry out experiments.
- Founded by Ex-Google CEO Eric Schmidt, Futurehouse says that AI supports real scientific breakthroughs, but details are kept on wraps.
- Lila Sciences, a startup with Now-Former Openai researcher Kenneth Stanley, has raised $200 million to create a lab dedicated to teaching AI models to ask more intelligent scientific questions, a fundamental prerequisite for insights.
Experts say why “new insights” from AI may not be useful yet
Even if the hype begins, top AI minds are hesitant. Thomas Wolf, chief science officer for embracing faces, recently argued that existing models cannot raise true new questions, a prerequisite for a scientific breakthrough. Similarly, Kenneth Stanley himself acknowledges that the problem is not merely a computational horsepower problem, but rather “fundamentally difficult” to model what humans consider to be “interesting” or “meaning.”Creativity is not just a calculation, it is a field of intuition and judgment that AI is still behind. This means that even if the model is capable of generating new hypotheses, it remains unknown how useful, testable, or modified these insights are. In science, novelties alone are not enough. It must be executable and verifiable. Until AI fills that gap, its “insights” can remain intellectually impressive, but essentially inert.
