Walter Laurito

Stephanie Milani, Anssi Kanervisto, Karolis Ramanauskas et al. · berkeley

To facilitate research in the direction of fine-tuning foundation models from human feedback, we held the MineRL BASALT Competition on Fine-Tuning from Human Feedback at NeurIPS 2022. The BASALT challenge asks teams to compete to develop algorithms to solve tasks with hard-to-specify reward functions in Minecraft. Through this competition, we aimed to promote the development of algorithms that use human feedback as channels to learn the desired behavior. We describe the competition and provide an overview of the top solutions. We conclude by discussing the impact of the competition and future directions for improvement.

Nicholas Popovic, Walter Laurito, Michael Färber

In this paper, we present an end-to-end joint entity and relation extraction approach based on transformer-based language models. We apply the model to the task of linking mathematical symbols to their descriptions in LaTeX documents. In contrast to existing approaches, which perform entity and relation extraction in sequence, our system incorporates information from relation extraction into entity extraction. This means that the system can be trained even on data sets where only a subset of all valid entity spans is annotated. We provide an extensive evaluation of the proposed system and its strengths and weaknesses. Our approach, which can be scaled dynamically in computational complexity at inference time, produces predictions with high precision and reaches 3rd place in the leaderboard of SemEval-2022 Task 12. For inputs in the domain of physics and math, it achieves high relation extraction macro F1 scores of 95.43% and 79.17%, respectively. The code used for training and evaluating our models is available at: https://github.com/nicpopovic/RE1st

Walter Laurito, Benjamin Davis, Peli Grietzer et al.

Are large language models (LLMs) biased in favor of communications produced by LLMs, leading to possible antihuman discrimination? Using a classical experimental design inspired by employment discrimination studies, we tested widely used LLMs, including GPT-3.5, GPT-4 and a selection of recent open-weight models in binary choice scenarios. These involved LLM-based assistants selecting between goods (the goods we study include consumer products, academic papers, and film-viewings) described either by humans or LLMs. Our results show a consistent tendency for LLM-based AIs to prefer LLM-presented options. This suggests the possibility of future AI systems implicitly discriminating against humans as a class, giving AI agents and AI-assisted humans an unfair advantage.

Walter Laurito, Sharan Maiya, Grégoire Dhimoïla et al.

The deployment of language models brings challenges in generating reliable information, especially when these models are fine-tuned using human preferences. To extract encoded knowledge without (potentially) biased human labels, unsupervised probing techniques like Contrast-Consistent Search (CCS) have been developed (Burns et al., 2022). However, salient but unrelated features in a given dataset can mislead these probes (Farquhar et al., 2023). Addressing this, we propose a cluster normalization method to minimize the impact of such features by clustering and normalizing activations of contrast pairs before applying unsupervised probing techniques. While this approach does not address the issue of differentiating between knowledge in general and simulated knowledge - a major issue in the literature of latent knowledge elicitation (Christiano et al., 2021) - it significantly improves the ability of unsupervised probes to identify the intended knowledge amidst distractions.

Kieron Kretschmar, Walter Laurito, Sharan Maiya et al.

Prior work has introduced techniques for detecting when large language models (LLMs) lie, that is, generating statements they believe are false. However, these techniques are typically validated in narrow settings that do not capture the diverse lies LLMs can generate. We introduce LIARS' BENCH, a testbed consisting of 72,863 examples of lies and honest responses generated by four open-weight models across seven datasets. Our settings capture qualitatively different types of lies and vary along two dimensions: the model's reason for lying and the object of belief targeted by the lie. Evaluating three black- and white-box lie detection techniques on LIARS' BENCH, we find that existing techniques systematically fail to identify certain types of lies, especially in settings where it's not possible to determine whether the model lied from the transcript alone. Overall, LIARS' BENCH reveals limitations in prior techniques and provides a practical testbed for guiding progress in lie detection.