@inproceedings{xia-etal-2026-ai,
title = "From {AI} Assistant to {AI} Scientist: Autonomous Discovery of {LLM}-{RL} Algorithms with {LLM} Agents",
author = "Xia, Sirui and
Zhang, Yikai and
Chen, Aili and
Wu, Siye and
Yuan, Siyu and
Xiao, Yanghua",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.2446/",
pages = "52498--52523",
ISBN = "979-8-89176-462-0",
abstract = "Discovering improved policy optimization algorithms for language models remains a costly manual process requiring repeated mechanism-level modification and validation. Unlike simple combinatorial code search, this problem requires searching over algorithmic mechanisms tightly coupled with training dynamics while reusing empirical evidence across iterations. We propose POISE a closed-loop framework for automated discovery of policy optimization algorithms for language models. POISE maintains an archive that links each candidate design with its implementation, evaluation evidence, and analysis summary, enabling later iterations to build on both successful and failed designs. In mathematical reasoning experiments starting from GRPO, POISE evaluates 64 candidate algorithms and discovers improved mechanisms, including analytic-variance scaling and validity masking. The best variant improves weighted Overall from 47.8 to 52.5 (+4.6) and increases AIME25 pass@32 from 26.7{\%} to 43.3{\%}, demonstrating the feasibility of automated policy optimization discovery while supporting interpretable design principles."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="xia-etal-2026-ai">
<titleInfo>
<title>From AI Assistant to AI Scientist: Autonomous Discovery of LLM-RL Algorithms with LLM Agents</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sirui</namePart>
<namePart type="family">Xia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yikai</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aili</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siye</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siyu</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanghua</namePart>
<namePart type="family">Xiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Discovering improved policy optimization algorithms for language models remains a costly manual process requiring repeated mechanism-level modification and validation. Unlike simple combinatorial code search, this problem requires searching over algorithmic mechanisms tightly coupled with training dynamics while reusing empirical evidence across iterations. We propose POISE a closed-loop framework for automated discovery of policy optimization algorithms for language models. POISE maintains an archive that links each candidate design with its implementation, evaluation evidence, and analysis summary, enabling later iterations to build on both successful and failed designs. In mathematical reasoning experiments starting from GRPO, POISE evaluates 64 candidate algorithms and discovers improved mechanisms, including analytic-variance scaling and validity masking. The best variant improves weighted Overall from 47.8 to 52.5 (+4.6) and increases AIME25 pass@32 from 26.7% to 43.3%, demonstrating the feasibility of automated policy optimization discovery while supporting interpretable design principles.</abstract>
<identifier type="citekey">xia-etal-2026-ai</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2446/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>52498</start>
<end>52523</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T From AI Assistant to AI Scientist: Autonomous Discovery of LLM-RL Algorithms with LLM Agents
%A Xia, Sirui
%A Zhang, Yikai
%A Chen, Aili
%A Wu, Siye
%A Yuan, Siyu
%A Xiao, Yanghua
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F xia-etal-2026-ai
%X Discovering improved policy optimization algorithms for language models remains a costly manual process requiring repeated mechanism-level modification and validation. Unlike simple combinatorial code search, this problem requires searching over algorithmic mechanisms tightly coupled with training dynamics while reusing empirical evidence across iterations. We propose POISE a closed-loop framework for automated discovery of policy optimization algorithms for language models. POISE maintains an archive that links each candidate design with its implementation, evaluation evidence, and analysis summary, enabling later iterations to build on both successful and failed designs. In mathematical reasoning experiments starting from GRPO, POISE evaluates 64 candidate algorithms and discovers improved mechanisms, including analytic-variance scaling and validity masking. The best variant improves weighted Overall from 47.8 to 52.5 (+4.6) and increases AIME25 pass@32 from 26.7% to 43.3%, demonstrating the feasibility of automated policy optimization discovery while supporting interpretable design principles.
%U https://aclanthology.org/2026.findings-emnlp.2446/
%P 52498-52523
Markdown (Informal)
[From AI Assistant to AI Scientist: Autonomous Discovery of LLM-RL Algorithms with LLM Agents](https://aclanthology.org/2026.findings-emnlp.2446/) (Xia et al., Findings 2026)
ACL