Optimize a specified algorithm
Real system improvements with automated verification; humans still define the objectives and evaluation functions.
SCIENCE UNDER AUTOMATION
AI in Algorithm design: capabilities and evidence
Up to 70% faster for length-5 sorting
What has AI demonstrated?
Up to 70% faster for length-5 sorting
- Test
- AlphaDev routines translated to C++; ~1.7% improvement for sequences over 250,000 elements.
- Comparison
- Sorting implementations in the C++ library used by the study.
- System and version
- AlphaDev · 2023
- Evidence checked
- 2026-10-20
Specific routines, hardware and workloads; not a 70% improvement across software development.
What does the field require?
Improve a specified executable routine
- Formulate objective and evaluator
Humans supply the optimization target and evaluation environment.
AlphaEvolve technical report - Search for executable improvements
Executable feedback supports iterative algorithm search.
AlphaEvolve technical report - Check correctness and performance
Automated tests cover the supplied evaluator, not arbitrary deployment conditions.
AlphaEvolve technical report
Solve a newly formulated algorithmic problem
- Choose the abstraction and constraints
No representative autonomous problem-formulation evaluation.
AlphaEvolve technical report - Develop an algorithm and proof
Search successes are narrower than unrestricted algorithm design.
AlphaEvolve technical report - Establish generalization and integrate
Independent end-to-end integration is not evaluated.
AlphaEvolve technical report
Research tasks
Can AI replace the whole workflow?
The results above measure specific tasks. The field assessment also requires completion of the connected workflow, including the requirements listed here. A field-wide replacement percentage has not been established.
Read the assessment method