How AI discovers better algorithms

THE QUESTION

How can AI find algorithms that improve on existing human-designed solutions?

WHY IT MATTERS

Algorithms determine how much work a computation needs and which problems become practical to solve.

THE ANSWER

AI can find better algorithms by proposing executable changes, testing what they do, and using the results to guide the next proposal. The useful output can be a better routine or a search program that uncovers a separate mathematical construction. Researchers can then simplify a discovered pattern and prove what it reveals about an existing algorithm. Whether a gain matters depends on testing the computation that will use it and on the value of studying or repeatedly using the result.

The evidence behind this report

We read 94 sources and cited 8. Every citation ships with a receipt — open one:

Read the full report →

The receipt for [1]

Academicarxiv.org

Novikov et al. AlphaEvolve: A coding agent for scientific and algorithmic discovery. arXiv:2506.13131v1 (2025), §§2–4 and 6; especially §3.1 for tensor…

Why we searched

How does AI develop a new algorithm through experimentation?

The cited claim

**Answer:** AI can find better algorithms by proposing executable changes, testing what they do, and using the results to guide the next proposal. [1]

How this source supports the claim

AlphaEvolve generates code modifications (diffs) via LLMs, applies them to create new programs, scores them with evaluators, and registers promising solutions back into the database to drive iterative discovery of better programs.

Cited in 14 places in this report

Click any marker — the receipt shows what we read and why we searched for it.

Recently published

All 36 reports →

19 Oct How World Models Turn Experience into Practice How learned models let agents turn past observations into new attempts—and turn practice into reusable skills. 9 cited · 82 read 17 Oct Can Tests Teach a Prompt to Improve? Automatic prompt optimization searches for instructions—and sometimes examples—that work better when executed. Its value depends on what the search changes, how it uses results, and what happens on cases kept outside the search. 10 cited · 96 read 14 Oct When is an agent's fast decision good enough? Additional reasoning is worth buying when a specific intervention can improve a consequential choice enough to justify its full cost. Confidence can help find those opportunities; it cannot price them by itself. 13 cited · 86 read 12 Oct How AI agents turn separate efforts into an accepted result See how useful AI teams connect responsibilities, evidence and acceptance—and judge their value by what one worker’s contribution lets another do and whether the resulting work survives the right checks. 16 cited · 148 read