Sitemap
A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.
Pages
Posts
Unified Multimodal Explanations Through Decomposition
Published:
Decomposition allows us to create unified pairwise explanations of a multimodal language model.
Interpreting Language Models Through Decomposition
Published:
By decomposing the hidden state of a language model, we are able to interpret why a prediction was made.
portfolio
publications
The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and Tool Calling: A Survey
Published in preprint, 2024
This survey paper examines the recent advancements in AI agent implementations, with a focus on their ability to achieve complex goals.
Recommended citation: Masterman, T., Besen, S., Sawtell, M., & Chao, A. (2024). The landscape of emerging ai agent architectures for reasoning, planning, and tool calling: A survey. arXiv preprint arXiv:2404.11584.
Download Paper
Lightweight Safety Classification Using Pruned Language Models
Published in preprint, 2024
This paper uses the intermediate hidden states of language models as features to train a simple logistic regression classifier. By doing so, our models surpass both fine-tuned models and general-purpose LLMs, reducing parameter count by up to 66%. Additionally, the classifiers can be trained using less than 100 examples without any modification of the underlying language model. This allows for many classification predictions to occur during the model’s forward pass with virtually zero computational overhead.
Recommended citation: Sawtell, M., Masterman, T., Besen, S., & Brown, J. (2024). Lightweight safety classification using pruned language models. arXiv preprint arXiv:2412.13435.
Download Paper