Spec-o3: A Tool-Augmented Vision-Language Agent for Rare Celestial Object Candidate Vetting via Automated Spectral Inspection

Minghui Jia, Qichao Zhang, Ali Luo, Linjing Li, Shuo Ye, Hailing Lu, Wen Hou, Dongbin Zhao · 2026

Read Spec-o3: A Tool-Augmented Vision-Language Agent for Rare Celestial Object Candidate Vetting via Automated Spectral Inspection by Minghui Jia, Qichao Zhang, Ali Luo, Linjing Li, Shuo Ye, Hailing Lu, Wen Hou, Dongbin Zhao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-06498.txt.

Description

Rare celestial objects still need a human expert's eyes before entering catalogs, and surveys now produce far too many candidates. Minghui Jia and 7 co-authors present Spec-o3 in a 2026 paper, a tool-augmented vision-language agent that inspects spectra the way astronomers do, using specialized tools and stepwise reasoning.

Training came in two stages: supervised fine-tuning on recorded expert inspection trajectories, then outcome-based reinforcement learning on rare-type verification tasks. On five rare-object tasks from the LAMOST survey, the agent raised the macro-F1 score dramatically, beating proprietary vision-language models and specialized deep classifiers with a modest base model. It generalized to unseen tasks and to other surveys. Experts rated its reasoning traces coherent and physically consistent, supporting trustworthy decisions.

Catalog record

CREATORMinghui Jia, Qichao Zhang, Ali Luo, Linjing Li, Shuo Ye, Hailing Lu, Wen Hou, Dongbin Zhao
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-06498

Browse the full library · About the Tower of Babel