AI Science Tracker
EN

Build, Don't Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players

Affiliation not stated in the reviewed material · 24 Oct 2026 · multidisciplinary

Discovery

Build, Don't Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players

Source check: Source checked. Not independent scientific confirmation.

Finding: Across 392 live ladder games, the account rating rose from 6,000 to 14,007; the paper separates 96 client failures from agent-play performance. The result is a single study account on a live game ladder; leader-board estimates and client failures are explicitly outside the bounded policy-performance claim.

Technology: LLM research agents improve the game simulator/opponents; a transformer policy trained with PPO chooses play actions.

Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.

AI contribution
LLM research agents improve the game simulator/opponents; a transformer policy trained with PPO chooses play actions.
Test and result
The result is a single study account on a live game ladder; leader-board estimates and client failures are explicitly outside the bounded policy-performance claim.
Limitation / still unknown
The reviewed source does not establish prospective benefit, broad external validity, or replication beyond the reported study.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/html/33845v1

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 22 Nov 2026. Review scope: Primary result-bearing material for the stated method, result and evaluation, with bibliographic version information..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse multidisciplinary discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.