Call for Papers
Significant advancements in Large Language Models (LLMs) have spurred interest in using them to assist researchers in various scientific tasks, such as idea generation, experiment execution and automated ML engineering, searching and synthesizing literature reviews and scientific theories, literature-based question-answering, data analysis and data-driven discovery, and even achieving novel state-of-the-art discoveries and executing the entire end-to-end research pipeline including paper and report generation.
These powerful AI systems are shown to be helpful tools for human researchers to accelerate the scientific discovery process across many scientific domains. For example, LLM-based agents have been developed for biologists to perform tasks such as data analysis and wet-lab assistance. We’ve also seen attempts at using LLMs for math research, such as research-level theorem proving and unsolved math problems. The paradigm of large-scale unified pretraining and reasoning post-training has been applied in various scientific domains, such as earth science, material science, and chemistry.
The focus of our workshop is on developing LLMs and AI systems that can accelerate scientific research in various scientific domains and assist human researchers. We aim to bring together researchers from different communities, including ML, NLP, Human-Computer Interaction, and various scientific disciplines such as biology and chemistry, to work together on the design, development, and evaluation of various forms of scientific LLMs and systems. Through our interdisciplinary program, we aim to foster new connections for this emerging community.
Topics of Interest
Potential topics of our workshop include but are not limited to:
Methods and Modeling Techniques
Methods and modeling techniques that lead to better scientific models and agents. Examples include:
- Training specialized LLMs for scientific domains or applications
- Developing LLM agents to accelerate complex scientific workflows
- New human-AI collaboration paradigms to empower human scientists
- Domain-specific foundation models for scientific applications
- Quantitative, symbolic, and physics-based reasoning in foundation models
- End-to-end research assistants and agents for scientific workflows
Datasets, Benchmarks, and Evaluations
Datasets, benchmarks, and evaluations for scientific AI systems and human-AI collaborations. Examples include:
- Careful evaluation and analysis of AI-generated research artifacts
- New automatic benchmarks and evaluators for measuring domain knowledge and capabilities of scientific LLMs
- Attribution, factuality, and hallucination mitigation in scientific contexts
- Multimodal scientific document understanding and knowledge extraction
Position Papers
Position papers on any related topics, such as:
- Societal impact and ethical concerns of deploying scientific AI systems
- Potential safety risks and mitigation for autonomous research agents
- Perspectives on the evolving role of AI in scientific discovery
Submission Guidelines
Our workshop will solicit submissions in full paper format (up to 8 pages of main content). All submissions should follow the COLM conference template. We will use OpenReview for submission management to ensure transparency and facilitate discussion.
Submit your paper here: openreview.net/group?id=colmweb.org/COLM/2026/Workshop/LM4Sci
The reviewing procedure will include a double-blind peer review with at least three reviewers per paper. Submissions will be non-archival, though accepted papers can be made available on the workshop website.
Important Dates
- Submission Deadline: June 23, 2026, 11:59pm AOE
- Reviews Due: July 20, 2026 AOE
- Decision Notification: July 24, 2026 AOE