AI-Researcher: an autonomous system for end-to-end AI research with Scientist-Bench

The authors present AI-Researcher, a system described as orchestrating literature review, hypothesis generation, algorithm implementation and manuscript preparation with minimal human intervention, alongside Scientist-Bench.

Workflow level n/a Source Preprint (not peer-reviewed) Outcome Author-reported

At a glance

Type

Framework

Workflow level

Workflow unspecified / not applicable

Year

2025

Disciplines

Computer science

Tasks

Hypothesis generationLiterature researchResearch codingScientific writing

Entry updated

2026-10-17

Evidence & verification

Source level

Preprint (not peer-reviewed)
Grounded in a preprint that had not completed peer review at the time the source was checked.

Outcome verification

Author-reported
Outcomes are reported by the team that built the system. The entry is source-checked against the team's own publications; not independently reproduced.

Sources checked

2026-10-17 (UTC)

Abstract page (including comments, subject class and submission history) read on 2026-10-17; raw HTML retained with SHA-256.

Source-checked does not mean independently reproduced. See the Method page for how these labels are assigned.

Sources (1)

Research applicability

Relevant to designing and evaluating computational research agents; benchmark effectiveness and scientific validity require separate assessment.

Getting started

Editorial suggestion, not exercised: start with the arXiv record and its linked HKUDS/AI-Researcher repository, then inspect setup and evaluation requirements before attempting a pilot. Researchers should verify citations, code and conclusions.

Maintenance

Version 1 only at retrieval time; no later revision observed on the page.

License

Preprint; arXiv licence terms apply, not verified in detail.

Workflow & human oversight

The authors describe a complete research pipeline; detailed iteration and human-oversight boundaries were not inspected, so workflow level remains unspecified.

Workflow level is not an evidence score; higher is not better evidence.

Approach

The abstract describes Scientist-Bench as comprising papers across AI research domains with guided-innovation and open-ended exploration tasks. Agent architecture, prompts and evaluation protocol were not inspected.

Limitations

Abstract-only evidence. Benchmark construction, baselines and the quality rubric were not inspected, and the reported success rates are author-reported rather than replicated.

Reproducibility

No artefact inspected beyond the abstract page; results were not reproduced here.

Adoption requirements

The abstract page links code, but hardware, model access, costs, privacy and setup requirements were not reviewed.

Comments

No comments yet.

Add a comment

Comments are held for moderation before appearing. Critical or corrective comments are welcome — especially evidence that an entry overstates what has been verified.