position-paper

This position paper argues against the claim that natural language can fully replace formal languages such as programming languages, proposing an information-theoretic specificity framework and proving a crossover theorem showing formal languages are better for high-specificity tasks.

This position paper argues for shifting from reactive AI flywheel maintenance (patching errors as they occur) to a proactive test-driven approach that maps feedback to a test space of task conditions, showing mathematically that the proactive method achieves better long-term scaling with fewer iterations.

This position paper argues for theory-level autoformalization, which formalizes entire theories including axioms, definitions, and lemmas as coherent libraries, rather than isolated statements. It discusses the significance, alternative views, open challenges, and proposes paths forward for this shift in formalization research.

This position paper argues that ground truth datasets in machine learning are not objective truths but human constructions shaped by choices, and advocates for articulating these choices to improve reliability, transparency, and accountability.

This position paper reviews the current state of LLM-driven formal mathematics, identifies key limitations in applying these systems to open-ended research mathematics, and proposes a strategic roadmap for developing AI agents capable of advancing mathematical frontiers.

This position paper argues that RL researchers need to distinguish between solving simulators and using simulators as a proxy for real deployment, highlighting issues that arise when this distinction is not made.

This position paper argues that the term 'machine unlearning' is overused in LLM research, advocating for stricter terminology tied to dataset-defined deletion and retraining-equivalence guarantees.

A position paper by Subbarao Kambhampati and researchers at Arizona State University argues that chain-of-thought reasoning in LLMs creates an illusion of reasoning, and the industry needs to move beyond costly token generation to alternative reasoning mechanisms.

This ICML 2026 spotlight position paper identifies a failure mode in image-generation alignment where aesthetic preference optimization overrides explicit user intent, terming it 'reversed alignment' and testing on anti-aesthetic prompts.

This position paper argues that current AI paradigms are insufficient for addressing global systemic risks and proposes Planet-Centered AI (PCAI) as a new design philosophy that treats Earth's interconnected systems as first-class concerns.

This position paper argues that integrating explicit memory, analogous to human hippocampal memory, is essential for advancing LLMs toward AGI. It draws on neuroscience to propose that higher-order cognitive functions require explicit memory beyond implicit statistical learning.

This position paper argues that current methods for evaluating AI resource usage are insufficient and advocates for the adoption of life cycle assessment (LCA) to properly account for energy and environmental costs across the entire ML pipeline, from hardware manufacturing to training and inference.

This position paper argues that large language models should learn from personalized rather than aggregated human preferences, highlighting theoretical limitations from social choice theory and practical issues from demographic diversity. It proposes bounded personalization frameworks that respect individual autonomy while maintaining universal safety constraints.

This position paper argues that a scientific understanding of AI must go beyond post-hoc analysis and instead study the training dynamics that shape model behavior, with implications for predicting, intervening, and designing training procedures for desired properties like capabilities and safety.

This position paper argues that deployed RL agents should never stop learning, as the train-then-fix paradigm inherently fails to address non-stationarity and distribution shift in real-world environments. The authors identify four sources of post-deployment non-stationarity and advocate for continual RL as the standard approach for deployed systems.

This position paper argues that sampling-based inference in Bayesian neural networks has achieved computational parity with optimization-based methods and is poised to supersede them, offering superior uncertainty quantification and prediction performance.

This position paper advocates for developing 'data probes'—synthetic sequences from random processes—to systematically study how data characteristics affect LLM performance, aiming to move beyond empirical heuristics.

This position paper argues that current uncertainty quantification methods for large language models are essentially unsupervised clustering, measuring internal consistency rather than external correctness, and therefore fail to detect confident hallucinations. The authors advocate for a paradigm shift to ground uncertainty in objective truth.

This position paper argues that machine learning research should prioritize ideas over benchmarks and theoretical guarantees, proposing an 'Ideas First' framework that values behavioral signatures and tailored experiments to promote equity and scientific understanding.

This position paper argues that incorporating metacognition as a design principle can lead to more accurate, secure, and efficient AI systems, and demonstrates the concept through a Federated Learning case study and a software framework for experimentation.