Microsoft Research released EvoLib on July 30th, a test-time learning framework that enables large language model agents to improve across tasks without updating their underlying weights. The framewor…
A new method called Lookahead-Enhanced Atomic Decomposition (LEAD) breaks the no-recovery bottleneck in long-horizon reasoning for Large Language Models, enabling the o4-mini model to solve Checkers J…
OpenAI research published in September 2025 and appearing in Nature shows that training language models for accuracy incentivizes hallucinations because benchmarks reward confident answers and penaliz…
Researchers have demonstrated a new prompt injection attack against leading large language models (LLMs) that achieves up to 60% success by writing malicious instructions in the style of the model's o…
A new wave of long-horizon agent benchmarks, including Agents' Last Exam, SWE-Marathon, the Meta-Agent Challenge, and Arena's Agent Mode, reveals a fundamental trade-off between realism and verifiabil…
Researchers at the 64th Annual Meeting of the Association for Computational Linguistics introduced a dual-task paradigm combining arithmetic and sentence comprehension to study how language models all…
Researchers introduced PrologMCP, an open-source server that exposes Prolog as a stateful tool through the Model Context Protocol, enabling LLM agents to delegate deductive reasoning tasks to a symbol…