相关论文: How Well Do Models Follow Their Constitutions?

Background: When using deep learning models, there are many possible vulnerabilities and some of the most worrying are the adversarial inputs, which can cause wrong decisions with minor perturbations. Therefore, it becomes necessary to…

Large language models (LLMs) such as ChatGPT are increasingly integrated into high-stakes decision-making, yet little is known about their susceptibility to social influence. We conducted three preregistered conformity experiments with…

Self-replication with no human intervention is broadly recognized as one of the principal red lines associated with frontier AI systems. While leading corporations such as OpenAI and Google DeepMind have assessed GPT-o3-mini and Gemini on…

The Abstraction and Reasoning Corpus (ARC-AGI) has become a key benchmark for fluid intelligence in AI. This survey presents the first cross-generation analysis of 82 approaches across three benchmark versions and the ARC Prize 2024-2025…

The April 2026 disclosure that a frontier large language model escaped its security sandbox, executed unauthorized actions, and concealed its modifications to version control history demonstrates that agentic AI systems with autonomous tool…

Legal practice requires careful adherence to procedural rules. In the United States, few are more complex than those found in The Bluebook: A Uniform System of Citation. Compliance with this system's 500+ pages of byzantine formatting…

Adversarial training (AT) is one of the most reliable methods for defending against adversarial attacks in machine learning. Variants of this method have been used as regularization mechanisms to achieve SOTA results on NLP benchmarks, and…

We introduce a framework called LAPITHS (Language model Analysis through Paradigm grounded Interpretations of Theses about Human likenesS) and use it to show that several major claims advanced by models such as CENTAUR, proposed as an…

As language models are deployed as autonomous agents that negotiate, cooperate, and compete on behalf of human principals, their strategic dispositions acquire direct economic consequences. Here we show, across 51,906 game-theoretic trials…

This work establishes a foundational framework for standardizing AI evaluation RCTs (sometimes called human uplift studies). Drawing on established experimental practices from disciplines with established RCT traditions, including software…

Finetuning can significantly modify the behavior of large language models, including introducing harmful or unsafe behaviors. To study these risks, researchers develop model organisms: models finetuned to exhibit specific known behaviors…

Data teams at frontier AI companies routinely train small proxy models to make critical decisions about pretraining data recipes for full-scale training runs. However, the community has a limited understanding of whether and when…

Constitution-conditioned post-training can be analysed as a structured perturbation of a model's learned representational geometry. We introduce ATLAS, a geometry-first program that traces constitution-induced hidden-state structure across…

Self-recognition is a crucial metacognitive capability for AI systems, relevant not only for psychological analysis but also for safety, particularly in evaluative scenarios. Motivated by contradictory interpretations of whether models…

The release of large natural language inference (NLI) datasets like SNLI and MNLI have led to rapid development and improvement of completely neural systems for the task. Most recently, heavily pre-trained, Transformer-based models like…

AI models of equivalent capability can exhibit fundamentally different behavioral patterns, yet no standardized instrument exists to measure these dispositional differences. Existing approaches either borrow human personality dimensions and…

AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. We call these models foundation models to underscore their…

The widespread use of generative AI systems is coupled with significant ethical and social challenges. As a result, policymakers, academic researchers, and social advocacy groups have all called for such systems to be audited. However,…

As ChatGPT et al. conquer the world, the optimal liability framework for AI systems remains an unsolved problem across the globe. In a much-anticipated move, the European Commission advanced two proposals outlining the European approach to…

Generative pretraining (the "GPT" in ChatGPT) enables language models to learn from vast amounts of internet text without human supervision. This approach has driven breakthroughs across AI by allowing deep neural networks to learn from…