aifollow.news
Back Rohan Paul
Rohan Paul· @rohanpaul_ai · X· · Original publication time AI score54

Huawei paper proposes a three-way division of labor for program improvementMachine translation

AI introduction

EvoTrias assigns search planning, code writing and scoring to separate agents, and keeps the evaluator hidden from the coding agent. It led on 6 of 7 math tasks and used 64% fewer tokens per run than four-agent CORAL. These results concern code optimization tasks, not general programming ability.

Original post · Original

The article text is unavailable in this language; an existing version is shown.

@rohanpaul_ai

– https://t.co/HAaFbtSrck

Title: "Where Autonomy Belongs: A Separation of Powers for Agentic Program Evolution"

ReplyRohan Paul@rohanpaul_ai
New Huawei paper finds that code-evolution agents do better and spend less when search planning, code writing, and scoring are handled by 3 separate owners. AlphaEvolve-style systems improve programs by writing code, scoring it, and picking what to try next. Some give the code writer little freedom, others let 1 coding agent steer the search at high token cost, and none hides the scorer. EvoTrias gives each job its own owner. A controller steers the search, a coding agent like Claude Code writes code, and an evaluator the agent cannot read scores it. On 7 math tasks it led 6 and cut token cost per run by 64% versus 4-agent CORAL. On OR-Tools, a 680K-line C++ repository, it lifted solver quality 13.2% on 300 instances, while 1-shot runs of the same agent averaged -0.1%. For code-optimizing agent loops, hide the scorer from the agent and keep search planning out of the coding agent.
View this post on X ↗

Source: Rohan Paul · x.com

Research
Found an error? Send a correction