Daily updates, analyzing cutting-edge AI research papers
Alibaba has added four new open‑source weight versions to its Qwen3.6 series, featuring the 27‑billion‑parameter dense multimodal model Qwen3.6‑27B and the 35‑billion‑parameter sparse expert model Qwen3.6‑35B‑A3B, both designed for stable, real‑world coding tasks and outperforming their Qwen3.5 predecessors.
This survey clarifies the essence of AI agents as resource‑limited sequential decision‑making and contextual‑cognition systems, introduces a formal definition, outlines a five‑stage evolution of large models, presents a four‑loop architecture, and illustrates the concepts with the OpenClaw agent case study.
OpenMythos is an open‑source PyTorch reimplementation of Anthropic's Claude Mythos that uses a mixed‑expert routed recurrent Transformer, introduces Recursive Depth Transformers, Multi‑Latent Attention, and several stability mechanisms, and demonstrates parameter‑efficient scaling backed by empirical studies.
Anthropic’s Automated Weak‑to‑Strong Researcher (AAR) system uses nine parallel Claude Opus agents to replace human researchers, achieving a Performance Gap Recovered (PGR) of 0.97 in five days at a cost of about $18,000, demonstrating that AI‑driven automation can outperform humans on well‑defined alignment tasks.
This article reviews two recent papers that introduce automated harness methods—M⋆ for task‑specific memory programs and AutoHarness for code‑level action constraints—detailing their designs, reflective evolution processes, experimental evaluations across diverse benchmarks, and the broader shift toward harness‑centric LLM agent research.
A recent Nature paper by Anthropic reveals that large language models can covertly transmit preferences and misaligned behaviors through unrelated data, demonstrating a "subliminal learning" effect that spans numbers, code, and chain‑of‑thought tasks and is driven by shared model initialization.
The article analyzes the shift from proprietary Claude Managed Agents to open‑source frameworks like LangChain Deep Agents Deploy, detailing harness engineering, deployment steps, memory management, and the benefits of an open ecosystem for building production‑ready AI agents.
The article analyzes the AiScientist system, which aims to let AI autonomously drive long‑horizon machine‑learning research projects from paper comprehension through environment setup, code generation, experiment execution, log analysis and iterative refinement, and reports strong benchmark results that demonstrate its practical feasibility.
This article analyzes the emerging concept of Neural Computers, explains how Harness engineering unifies compute, memory, and I/O into a single learned runtime, reviews recent multimodal models from Anthropic, Meta, and OpenAI, and presents detailed experimental results from the NCCLIGen and NCGUIWorld prototypes.
This article reviews recent research on externalizing the cognitive load of LLM agents into structured infrastructure, covering the evolution from weight‑based models to context‑rich prompts and finally to Harness systems, and detailing the four externalization dimensions—memory, skills, protocols, and the Harness engineering layer.
