About
Research & Development
Unpacking ‘Optimized Fragility’ in AI Models with In-Context Learning
TLDR: Research on GPT-OSS:20b models shows that In-Context Learning (ICL) guides, while improving efficiency and accuracy on general knowledge tasks (91-99% accuracy), lead to...
Research & Development
AI Agents Struggle in Hostile Financial Markets: A New Benchmark Reveals Critical Flaws
TLDR: A new benchmark, CAIA, evaluates AI agents in adversarial, high-stakes cryptocurrency markets, revealing that even advanced models achieve low accuracy (12-28% without tools,...
Research & Development
New Defense Against Deepfakes: Disrupting Diffusion Model Editing
TLDR: The research introduces DDIM Inversion Attack (DIA), a novel defense mechanism against malicious image editing using diffusion models. Unlike previous methods, DIA directly...
Research & Development
Unifying Semantics: Bridging Cost-Based Reasoning and c-Representations in Knowledge Systems
TLDR: This research paper compares two distinct approaches for handling uncertainty and inconsistency in Description Logics: cost-based semantics and c-representations. Both frameworks assign numerical...
Research & Development
Unlocking New Relations in Multimodal Knowledge Graphs with FusionAdapter
TLDR: FusionAdapter is a novel framework designed to improve few-shot relation learning in Multimodal Knowledge Graphs (MMKGs). It addresses the challenges of integrating diverse...
Research & Development
Decoding Human-Like Thinking in Large Language Models
TLDR: A study evaluated 77 large language models on a human-like conceptual representation task, finding that instruction-finetuning and larger architectural dimensions significantly improve alignment...