Researchers have explored the use of open-weight large language models (LLMs) as acquisition policies for materials optimization tasks. In a controlled study, five LLMs were evaluated across four retrospective finite-pool materials optimization problems. The LLMs generally outperformed random selection in finding optimal materials with fewer iterations, suggesting they can provide a useful signal without task-specific training. However, their performance relative to traditional Gaussian-process methods was mixed, with LLMs matching or exceeding Gaussian processes in some scenarios but underperforming in others. The study found that LLM performance varied significantly based on the specific task, the model used, and how candidate materials and scientific context were presented. AI
IMPACT LLMs may offer a new, generalizable approach to accelerating materials discovery, potentially reducing research costs and timelines.
RANK_REASON The cluster contains a research paper published on arXiv detailing a study on LLMs for materials optimization. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- CatalyzeX
- DagsHub
- Florian Le Bronnec
- Gaussian process
- Gotit.pub
- Hugging Face
- Large language models
- Materials Optimization
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
