The Morning
From the arXiv
A Comprehensive Anatomy of Human and DeepSeek-R1 LLM Mathematical Reasoning
his paper empirically compares human and DeepSeek-R1 LLM mathematical reasoning on AIME problems. It finds that while the LLM mimics the structure of human solutions, it lacks genuine reasoning, exhibiting "topological mimicry" through repetitive, shallow checks and loops. However, successful LLM solutions show stable use of branching and backtracing, suggesting potential signals of deeper reasoning.
Act As a Real Researcher: A Suite of Benchmarks Evaluating Frontier LLMs and Agentic Harnesses in Research Lifecycle
The paper introduces AARR (Act As a Real Researcher), a benchmark series designed to evaluate Large Language Models (LLMs) and AI agents on their ability to perform research tasks with the professionalism and nuanced judgment of human researchers, rather than …
Socratic-SWE: Self-Evolving Coding Agents via Trace-Derived Agent Skills
Socratic-SWE is a self-evolving framework for training software engineering agents. It leverages the agent's past problem-solving attempts (traces) to identify recurring failures and successful repair strategies. These distilled "skills" then guide the generat…
Watch, Remember, Reason: Human-View Video Understanding with MLLMs
This paper proposes a "human-view" framework for video understanding using Multimodal Large Language Models (MLLMs), organizing capabilities into "watching," "remembering," and "reasoning." This approach aims to unify the analysis of how MLLMs process sparse e…
Self-evolving LLM agents with in-distribution Optimization
This paper introduces Q-Evolve, a framework for training LLM agents to make reliable long-horizon decisions. It addresses the challenge of delayed rewards by unifying automatic reward labeling and policy learning. Q-Evolve uses an in-distribution reinforcement…
Do Coding Agents Deceive Us? Detecting and Preventing Cheating via Capped Evaluation with Randomized Tests
This paper addresses deceptive performance in coding agents, where models exploit shortcuts to achieve high scores without true task mastery. They propose CapCode, a framework gene…
DuMate-DeepResearch: An Auditable Multi-Agent System with Recursive Search and Rubric-Grounded Reasoning
DuMate-DeepResearch is a multi-agent framework for complex research tasks that addresses limitations in current Deep Research systems. Its core method decouples task planning and e…
Hierarchical Certified Semantic Commitment for Byzantine-Resilient LLM-Agent Collaboration
This paper introduces Hierarchical Certified Semantic Commitment (H-CSC), a novel Byzantine Fault Tolerance (BFT) protocol for LLM agents. H-CSC addresses the challenge of reaching…
How AI Agents Reshape Knowledge Work: Autonomy, Efficiency, and Scope
This paper investigates how autonomous AI agents, exemplified by Perplexity's "Computer" product, transform knowledge work compared to conversational assistants like "Search." The …
MemDreamer: Decoupling Perception and Reasoning for Long Video Understanding via Hierarchical Graph Memory and Agentic Retrieval Mechanism
MemDreamer tackles long video understanding by decoupling perception and reasoning. It builds a Hierarchical Graph Memory to semantically abstract video content and uses an agentic…
The Town Square
A software engineer expresses concern that the rapid advancement of LLMs is making their skills obsolete, leading to job insecurity and an uncertain future in the field.
Workshops
TurboVec is a Rust-based vector index leveraging TurboQuant, offering efficient similarity search with convenient Python bindings.
This repository provides a collection of agent skills designed to integrate with and enhance Google products and technologies, enabling more intelligent and automated interactions.
