GPT-Micro: A large language paradigm for accelerated, inexpensive, and thermodynamics-consistent discovery of constitutive models in manufacturing
Soumik Dutta, Kiarash Naghavi Khanghah, Sania Shree, Logan McNeil, Thomas Feldhausen, Hongyi Xu, Rajiv Malhotra
Abstract
Constitutive modeling of the relationship between process-imposed material states and fundamental material properties is critical to control of material microstructure in manufacturing processes. The limited accuracy resulting from the typical reliance on fallible human expertise and intuition for postulation and revision of the models functional form results in incremental and time consuming model discovery. Conventional Machine Learning (ML) incurs significant cost and time of data generation. Model discovery using Large Language Models (LLMs) suffers from the above issues and/or ignores the inviolability of fundamental thermodynamics laws. This work creates a novel GPT-Micro paradigm for autonomous, data sparse, and thermodynamics-compliant discovery of de-novo constitutive models. This framework seamlessly integrates semantic knowledge extraction from literature, enforcement of thermodynamics-based conservation laws, and sparse datasets, with LLM-driven generation and refinement of model hypotheses. Validation is performed for a long-intractable constitutive modeling problem in a printed electronics process testbed. This reveals significant and simultaneous advantages over the state-of-the-art including: (a) More than 70 percent reduction in data burden relative to ML-based modeling without loss in accuracy; (b) 400X reduction in discovery time after data generation, from months to hours, relative to human-driven modeling; (c) Discovery of models with novel functional forms without subjective human choice of a starting hypothesis; (d) Enhanced physics-rooted trustworthiness, human interpretability, and mechanistic insight via synthesis of compact, conservation-compliant, and physically complete analytical models. The potential of GPT-Micro to realize rapid, low-cost, physically trustworthy, and interpretable microstructure modeling across the manufacturing landscape is discussed.
AI Impact Assessments
(1 models)Scientific Impact Assessment: GPT-Micro
1. Core Contribution
GPT-Micro proposes a pipeline for autonomous discovery of constitutive models (relating process-imposed material states to material properties) by integrating: (a) RAG-based knowledge extraction from literature, (b) LLM-driven iterative hypothesis generation and refinement for state-microstructure models, (c) enforcement of thermodynamic conservation laws to generate synthetic data, and (d) symbolic regression to discover closed-form constitutive equations. The central insight is that LLMs can replace human intuition in postulating the functional form of constitutive models, while conservation laws can be enforced structurally through the pipeline rather than as soft constraints. The framework is validated on a nanowire sintering problem in printed electronics.
2. Methodological Rigor
Strengths in methodology:
Weaknesses and concerns:
3. Potential Impact
The paper addresses a genuine pain point: constitutive model discovery is slow, expertise-dependent, and often the bottleneck in computational manufacturing. If GPT-Micro generalizes, it could significantly accelerate the adoption of computational modeling for new materials and processes. Specific impact vectors include:
However, the impact is tempered by the single-problem validation and the requirement that conservation laws must be expressible in forms amenable to the pipeline (algebraically solvable for material properties given state-microstructure model outputs).
4. Timeliness & Relevance
The paper is highly timely. The convergence of LLM capabilities, the push for physics-informed ML, and the need for rapid constitutive modeling in advanced manufacturing creates a receptive audience. The integration of RAG with scientific hypothesis generation aligns with the broader "AI for Science" movement. The emphasis on thermodynamic consistency addresses a legitimate criticism of pure data-driven approaches.
5. Strengths & Limitations
Key Strengths:
Notable Weaknesses:
6. Additional Observations
The framework's reliance on GPT-4o-mini introduces dependency on a commercial API with no guarantees of reproducibility over time. The claim of "autonomy" is somewhat overstated — the user must supply context keywords, conservation law equations, R² thresholds, and the general mathematical framework. The true novelty lies in automating the hypothesis generation/refinement loop, not in full autonomy.
The paper would benefit significantly from validation on at least 2-3 additional manufacturing problems with different conservation law structures and dimensionalities to substantiate its generalizability claims.
Generated Jun 9, 2026
Comparison History (19)
Paper 2 tackles a fundamental and widespread problem in deep learning (loss of plasticity in continual learning) using rigorous theoretical foundations (dynamical isometry and NTK). Its proposed solutions, including a new optimizer (AdamO), have the potential for broad adoption across numerous AI subfields like reinforcement learning and supervised learning. While Paper 1 is highly innovative and impactful within manufacturing and materials science, Paper 2's methodological advancements offer a wider breadth of impact across the entire machine learning community.
Paper 2 is a comprehensive review article that proposes novel organizing frameworks (phase diagram of discoverability, REO framework) for the rapidly growing field of data-driven equation discovery. Reviews in high-impact venues tend to be highly cited as they serve as reference points for entire fields. Its breadth across physics and adjacent sciences, combined with its forward-looking perspective on theory revision and mechanism discovery, gives it wider impact potential. Paper 1, while innovative in combining LLMs with thermodynamic constraints for constitutive modeling, addresses a more specialized manufacturing niche with narrower applicability.
GPT-Micro addresses a broader, cross-disciplinary challenge—autonomous discovery of constitutive models in manufacturing—by integrating LLMs with thermodynamic constraints and sparse data. Its novelty lies in combining semantic knowledge extraction, physics compliance, and LLM-driven hypothesis generation, offering 70% data reduction and 400X speedup. This paradigm has wide applicability across manufacturing and materials science. Paper 1, while technically strong with impressive compiler optimization results, addresses a narrower problem (tensor program search) with incremental improvements to existing auto-schedulers, limiting its broader scientific impact.
Paper 1 has higher likely cross-field scientific impact: it introduces a generally applicable causal-intervention framework for attributing failures in LLM agents, a rapidly growing and broadly relevant area (AI safety, debugging, evaluation, reliable autonomy). The methodological core (SCM framing, do-operator replay, confound handling, Shapley credit with CIs) is comparatively rigorous and reusable across domains and agent architectures. Paper 2 appears highly impactful within manufacturing/materials, but its scope is narrower and validation seems centered on a specific testbed, making generalization and broad uptake less certain.
GPT-Micro presents a more complete and validated framework with demonstrated quantitative advantages (70% data reduction, 400X time reduction) for a well-defined, high-impact problem in manufacturing. It integrates multiple innovations—LLM-driven model discovery, thermodynamic compliance, and sparse data utilization—into a novel paradigm with clear practical applications. Paper 2 is more of a proof-of-concept position paper outlining design properties and open challenges for future Graph Foundation Models, with narrower validation (only super-spreader identification) and less mature contributions. Paper 1's broader applicability to manufacturing and stronger empirical validation suggest higher near-term scientific impact.
GPT-Micro presents a more transformative paradigm with broader cross-disciplinary impact, combining LLMs with thermodynamics-compliant constitutive model discovery in manufacturing. It demonstrates dramatic quantitative improvements (70% data reduction, 400X time reduction) on a real-world problem, bridging AI with materials science and manufacturing. Paper 2, while technically solid, offers an incremental improvement to activation steering methods within the narrower LLM interpretability/control community. GPT-Micro's novelty in integrating physics constraints with LLM-driven scientific discovery addresses a more fundamental challenge with wider practical applications.
GPT-Micro demonstrates broader scientific impact by addressing a widely relevant problem—constitutive model discovery in manufacturing—with dramatic practical improvements (70% data reduction, 400X faster discovery) while maintaining thermodynamic consistency. Its framework combining LLMs with physics constraints is highly novel and applicable across manufacturing domains. Paper 2, while mathematically sophisticated in extending neural rough differential equations to Itô/manifold settings via Hopf algebras, addresses a more specialized audience in stochastic analysis and geometric deep learning. Paper 1's interdisciplinary reach, practical utility, and timeliness (leveraging LLMs for scientific discovery) give it higher potential impact.
Paper 2 demonstrates exceptionally high potential for real-world impact by successfully bridging LLMs, physical laws, and manufacturing. It addresses a critical bottleneck in materials science with massive quantifiable leaps: a 400x reduction in discovery time and a 70% decrease in data needs. While Paper 1 offers excellent foundational ML improvements for scaling neural fields, Paper 2's novel integration of strict thermodynamics constraints with generative AI offers a transformative 'AI for Science' paradigm shift with immediate, far-reaching industrial applications.
Paper 1 introduces a paradigm-shifting approach bridging LLMs, thermodynamics, and manufacturing. By enabling autonomous, physics-compliant discovery of constitutive models, it significantly reduces discovery time and data requirements. Its broad applicability across physical sciences represents a major leap in AI-driven scientific discovery. In contrast, Paper 2 offers a valuable but more incremental algorithmic improvement for a specific application (recommender systems), making its broader scientific and interdisciplinary impact relatively lower.
Paper 2 presents a novel paradigm (GPT-Micro) that bridges LLMs with physics-constrained constitutive modeling in manufacturing, demonstrating dramatic practical improvements (70% data reduction, 400X faster discovery). It addresses a broadly relevant problem across manufacturing sciences with immediate real-world applications. While Paper 1 makes rigorous theoretical contributions to understanding deep GP priors (sharp thresholds, non-Gaussian limits), its impact is more niche, primarily advancing theoretical understanding within the Bayesian deep learning community. Paper 2's combination of methodological novelty, practical utility, interdisciplinary reach, and timeliness (leveraging the LLM revolution for scientific discovery) gives it higher potential impact.