Dear Readers,

Three years ago, the iconic AI demo was a wobbly unicorn doodle; today, it’s a model tightening theorems with a Fields Medalist and proposing new experiments to immunologists staring at mysterious plots. Somewhere between those two images - the childish sketch and the crowded lab bench - science quietly crossed a threshold. This issue is about that moment. What does it actually mean when a general-purpose model like GPT-5 stops being a toy and starts behaving like a slightly erratic, frighteningly fast colleague at the research frontier? And what should we make of a roadmap where “intern-level AI scientist in 2026” and “fully automated researcher by 2028” are no longer sci-fi lines, but explicit corporate goals?

We’ll walk through the new GPT-5 science paper in detail: from tightening convex-optimization bounds and solving Erdős problems, to uncovering hidden symmetries in black-hole physics and generating genuinely new immunology hypotheses from unpublished flow-cytometry plots. We’ll look at where the system breaks, how much human scaffolding it still needs, and what an AI-native lab might actually look like when models are baked into every step, from literature search to experiment design. And in Chubby’s Opinion Corner, we’ll zoom out: is this still “just” acceleration, or the early rehearsal for a world where AI becomes a full research colleague? If you want to understand where the unicorn era ends and the new scientific age begins, this is the issue to read closely.

All the best,

“Science shapes everything from human health to energy production, from national security to our understanding of the universe. If AI can accelerate science, shortening the time it takes to generate new ideas, or to move from an idea to a tested result, the benefits compound across society.”

— OpenAI

”AI will contribute to the world in many ways, but the gains to quality of life from AI driving faster scientific progress and increased productivity will be enormous; the future can be vastly better than the present. Scientific progress is the biggest driver of overall progress; it’s hugely exciting to think about how much more we could have.”

— Sam Altman

When Sébastien Bubeck writes that three years ago the frontier demo for AI was “a unicorn drawing” and today it is a model helping to solve real-world research problems, he is putting a stake in the ground. Together with Sam Altman’s comment that this is “a first preview of something we expect to see a lot more of,” the new paper, Early Science Acceleration Experiments with GPT-5 is meant as more than marketing. It is a status report from the lab bench: here is what a general-purpose model can already do for working scientists - and here is where it still breaks.