Primary evidence
- Faster sorting algorithms discovered using deep reinforcement learningS1 · Primary research · Paper · 2023-06-07
- AlphaDev accompanying repositoryS1 · Primary research · Repository · 2023-06-07
AlphaDev discovers faster sorting algorithms with deep reinforcement learning
The system searches directly over assembly instructions using an AlphaZero-style learning and search procedure. DeepMind reported speedups up to 70% for short sequences and about 1.7% for very long sequences, while releasing pseudocode and discovered assembly programs with correctness tests.
What happened?
AlphaDev formulated low-level algorithm design as a reinforcement-learning game and discovered new sorting routines that outperformed prior human-designed benchmarks; fixed sort-3, sort-4 and sort-5 routines were integrated into the LLVM standard C++ library.
Why it matters
The result showed that AI-discovered algorithmic improvements could move from a research search space into widely used production software, connecting exact algorithm discovery with measurable real-world deployment.
Significance rationale: AlphaDev extended reinforcement-learning-based algorithm discovery from mathematical search benchmarks into low-level computer-science routines whose discovered sorting algorithms were integrated into LLVM libc++, providing a deployed example of AI-generated exact algorithms.
Historical context
Earlier developments
Related milestones
—
Evidence & status at a glance
- Significance
- H2 · Field milestone
- Verification
- Paper released
- Evidence level
- E3 · Inspectable artifact
- AI role
- AI primary, human verified
- Formal assurance
- Artifact available
Contribution & verification
- Mathematical novelty
- New algorithm
- Interface
- Symbolic, Numeric, Hybrid
- AI contribution
- AlphaDev searched the assembly-program space and discovered candidate sorting routines under correctness and latency rewards; the resulting algorithms were tested, translated and integrated by researchers and software maintainers.
- Human contribution
- Researchers formulated AssemblyGame, designed the reinforcement-learning system and evaluation, analyzed the discovered routines and translated suitable discoveries into C++ for library integration.
- Verification
- Paper released · E3 · Inspectable artifact · Artifact available
- Last reviewed
- 2026-09-11
Technical details
- Systems
- alphadev
- Organizations
- google-deepmind
- Problems
- sorting-algorithms
- Methods
- reinforcement-learning, alphazero, program-search, assembly-search, algorithm-discovery
Sources & artifacts
- Faster sorting algorithms discovered using deep reinforcement learningS1 · Primary research · Paper · accessed 2026-09-11
- AlphaDev discovers faster sorting algorithmsS2 · Official institutional · Official announcement · accessed 2026-09-11
- AlphaDev accompanying repositoryS1 · Primary research · Repository · accessed 2026-09-11
Verification history
- 2023-06-07 · Paper released · E2 · Research record
Nature published the AlphaDev algorithm-discovery results and Google DeepMind released an official technical account.
- 2023-06-07 · Paper released · E3 · Inspectable artifact
The public AlphaDev repository exposes agent pseudocode, discovered assembly programs and correctness tests. E3 records inspectability rather than independent verification of every performance claim.
Corrections
No corrections recorded.