How Long Is a Piece of String? A Brief Empirical Analysis of Tokenizers
This paper explores the significant variation in tokenization across models and domains of text, challenging common heuristics about token lengths and providing insights into tokenization in contemporary LLMs.
LLMs, You Can Evaluate It! Design of Multi-perspective Report Evaluation for Security Operation Centers
The paper introduces a novel framework for evaluating analysis reports in Security Operation Centers, leveraging LLMs for effective report evaluation and maximization of feedback quality.
Building Production-Ready Probes For Gemini
This research identifies challenges in generalizing activation probes for LLMs and proposes robust architectures to enhance their usage in production, showcasing promising results.
Meta-Learning Guided Pruning for Few-Shot Plant Pathology on Edge Devices
This work presents a novel framework combining pruning with meta-learning to improve edge deployment of models for plant disease classification, showing significant reductions in model size while maintaining accuracy.
Capacity Constraints Make Admissions Processes Less Predictable
The study reveals how admissions decision-making differs from conventional ML paradigms due to capacity constraints and highlights the implications for machine learning model performance.
Exploring LLM Features in Predictive Process Monitoring for Small-Scale Event-Logs
This paper extends an existing LLM-based predictive process monitoring framework, demonstrating improved predictions in low-data settings and showcasing the reasoning strategies employed by the model.
Health Facility Location in Ethiopia: Leveraging LLMs to Integrate Expert Knowledge into Algorithmic Planning
The study proposes a hybrid framework that combines expert knowledge with optimization techniques for health facility upgrades in Ethiopia, demonstrating effective solutions for equitable health system planning.
The Poisoned Apple Effect: Strategic Manipulation of Mediated Markets via Technology Expansion of AI Agents
This research examines how the introduction of new technologies can manipulate regulatory outcomes and emphasizes the need for dynamic market designs to counter potential exploitation.
Engineering Agency into Large Language Models: Phase I Results
This paper discusses the engineering of agency into large language models, showcasing preliminary results that improve their interactive capabilities through structured interventions.