TractorBeam suggests that systems that facilitate exploratory research on individual documents may lead to verifiable sensemaking for users and complement tools that work across broader corpora.
Abstract
Language model-based systems which allow asking questions of documents have become popular tools for sensemaking. Despite their implied capability, these systems still suffer from issues of factuality and provenance, while encouraging confirmatory, rather than exploratory, research. We present TractorBeam, a browser extension-based mixed-initiative system that uses collaborative annotation as an interface metaphor for sensemaking, re-framing language model (LM) outputs as suggested highlights in a process that we call collaborative machine annotation. This metaphor allows us to present LM results in-context on PDF documents, directly addressing concerns of provenance and factuality, while allowing users to iteratively construct mental schemas and queries for language models directly in the context of a document. In a preliminary user study, all of our participants felt that TractorBeam enabled them evaluate and iteratively improve the model's reflection of their intended highlighting, and several found suggestions that made them reconsider their original schema. TractorBeam suggests that systems that facilitate exploratory research on individual documents may lead to verifiable sensemaking for users and complement tools that work across broader corpora.
This work presents a series of architectural improvements and feature enhancements to AquiLLM, including local embedding and reranking, multimodal capabilities, OpenAI-compatible inference interfaces, user interface improvements, semantic and episodic memory capabilities, and skills support.
J. Stark, S. Saikrishnan, Vikram Seenivasan et al.· 1 citation
For decades, search and recommendation systems have been optimized as distinct components within large-scale discovery platforms. The rise of generative AI is beginning to blur this boundary. At Spotify, we are exploring how large language models can evolve from tools that retrieve content into systems that reason over...
Paul N. Bennett· Proceedings of the 32nd ACM...· 0 citations
Travel literature is a unique form of hypertext that extends beyond its medium into real-world physical exploration. While conventional computational methods can easily extract surface-level ontological entities (e.g., locations, dates), the deeper epistemic and judgmental subtext that guides the discovery and evaluati...
Virginia Orlando, Alessandro Adamou, Alessio Antonini· Proceedings of the 37th ACM...· 0 citations
Archives are essential vehicles for both historical research and memory practices like genealogy. The sheer volume of documents produced via digital infrastructures, however, has posed a challenge for traditional archival processes that rely on human judgments. Some archivists have responded to this challenge by en...
ReSB2 is introduced, a framework for retrieving and linking similar legislative bills that supports human–machine collaboration and helps reduce redundancy in the lawmaking process, and fine-tunes two ModernBERT-based language models on authentic legislative data.
Lucas G. L. Costa, Átila M. Souza, Elves Rodrigues et al.· Proceedings of the 37th ACM...· 0 citations
A multidimensional taxonomy is proposed to classify grounding techniques by architectural paradigm, data source, underlying mechanism, and intervention phase, indicating they are transitioning from an optional improvement to a highly prioritized architectural component for LLM-based recommender systems.
Andrés Felipe Solis Pino, N. Duque-Méndez, Pablo H. Ruiz et al.· Applied Sciences· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.