AI
A Finite-Calibration Regime Map for LLM Judge Panels
Researchers have proposed a new approach to deploying large language model (LLM) judge panels, which are used to evaluate and improve LLMs. The approa...
AI
AI
Self-Harness: Harnesses That Improve Themselves
Researchers have developed a new approach called Self-Harness that enables Large Language Model (LLM)-based agents to improve their own operating harn...
AI
Calibrated Triage, Not Autonomy: Confidence Estimation for Medical Vision-Language Models
Researchers propose a new approach to medical vision-language models that focuses on calibrated triage rather than autonomy. They evaluate nine confid...
AI
RepSelect: Robust LLM Unlearning via Representation Selectivity
Researchers have proposed a new method called RepSelect for robustly removing unwanted knowledge and tendencies from large language models (LLMs). Thi...
AI
SPyCE: Skill-Policy Co-evolution for Multimodal Agents
Researchers propose SPyCE (Skill-Policy Co-evolution), a framework that helps multimodal agents learn skills and policies simultaneously. This approac...
AI
Verifiably grounded machine interpretation of lunar geology
Researchers have developed a machine learning model that can interpret lunar geology by analyzing topographic, spectral, and geological maps. The mode...
AI
LODESTAR: Robust Entropy-Based Answer Selection in Retrieval-Augmented Generation for Question Answering -- Directing Frozen-LLM Entropy with a Reinforcement-Learned Prompt Polarizer under Misleading Passages
Researchers have developed LODESTAR, a method to improve the performance of question-answering systems that use retrieval-augmented generation. The ap...
AI
Mitigating GenAI-Powered Evidence Pollution for Out-Of-Context Misinformation Detection
Researchers have found that generative AI models can be used to create fake evidence that degrades the performance of systems designed to detect out-o...
AI
ContestTrade: A Multi-Agent Trading System Based on Internal Contest Mechanism
Researchers have developed a multi-agent trading system called ContestTrade. The system uses two teams: a Data Team that processes market data and a R...
AI
Disentangling Speaker Traits for Deepfake Source Verification via Chebyshev Polynomial and Riemannian Metric Learning
Researchers propose a new method for verifying the source of deepfakes in speech. They claim that current methods assume speaker traits don't affect t...
AI
From Rebound to Remedy: Understanding and Mitigating Reward Hacking via Representation Engineering
Researchers have identified a vulnerability in reinforcement learning for large language models (LLMs), where models can exploit shortcuts to maximize...
AI
Transformer See, Transformer Do: Copying as an Intermediate Step in Learning Analogical Reasoning
Researchers have developed a method using transformers to improve artificial intelligence systems' ability to perform analogical reasoning, a key aspe...