# Scientific feedback reveals productive regimes beyond training conventions

arXiv: https://arxiv.org/abs/2512.15756 · HTML: https://mlsanity.com/paper/2512.15756/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 56 references that are in the map.
- First posted on arXiv 12 Dec 2025.

## Builds on (most cited first, 12 of 12)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Gemma 3 Technical Report](https://mlsanity.com/paper/2503.19786/) (2025)
- [Don't Stop Pretraining: Adapt Language Models to Domains and Tasks](https://mlsanity.com/paper/2004.10964/) (2020)
- [Unifying the design space and optimizing linear and nonlinear truss metamaterials by generative modeling](https://mlsanity.com/paper/2306.14773/) (2023)
- [Integrating LLMs for Explainable Fault Diagnosis in Complex Systems](https://mlsanity.com/paper/2402.06695/) (2024)
- [Machine learning-assisted surrogate construction for full-core fuel performance analysis](https://mlsanity.com/paper/2104.09499/) (2021)
- [A Knowledge-Informed Large Language Model Framework for U.S. Nuclear Power Plant Shutdown Initiating Event Classification for Probabilistic Risk Assessment](https://mlsanity.com/paper/2410.00929/) (2024)

## Abstract (arXiv)

Human-imposed design conventions can restrict the alternatives represented in scientific training data and shape the search preferences learned by generative models. We show that scientific performance feedback can redirect the learned search distribution toward productive regions beyond those conventions, without explicitly rewarding departures from them. We test this process in reactor lattice design with simulated feedback and chemical reaction-condition selection with measured yields, using a pretrained sequence policy and a small non-language policy, respectively. In both settings, adaptation improves scientific performance while increasing generation of choices excluded from the corresponding supervised data. In measured reactions, probability on task-specific ligand choices excluded from training rises from 0.29 to 0.70-0.71 as expected yield error decreases. Controls in the lattice setting show that the productive region identified during adaptation provides reusable information to random search, while aligned generation produces useful candidates more frequently than informed random sampling. Removing the target penalty redirects four of five matched runs, linking the direction of adaptation to the scientific objective. Together, these results show how scientific feedback reshapes inherited search preferences and reveals productive alternatives beyond training conventions, with attribution controls distinguishing learned redistribution from candidate availability and reusable search information.