PinnedInTowards DevbyArXiv In-depth Analysis·Mar 26, 2025Beyond the Window: The Complete Guide to Long Context Language Models (And Why They’re Changing AI)How to obtain effective and efficient LCLMs, train and deploy LCLMs efficiently, evaluate and analyze LCLMs comprehensively.
PinnedInTowards DevbyArXiv In-depth Analysis·Mar 27, 2025Chain-of-Tools: Unleashing Frozen LLMs on a Universe of Unseen ToolsChain-of-Tools makes full use of the powerful semantic representation capability of frozen LLMs to finish tool calling in CoT reasoning
PinnedInTowards DevbyArXiv In-depth Analysis·Mar 29, 2025CodeAct: The Engine Behind Manus— How LLMs Are Learning to Code Their Way to ActionUnlocking Agent Potential: A Deep Dive into CodeAct, Python Actions, and the LangGraph-CodeAct Library.
PinnedInGoPenAIbyArXiv In-depth Analysis·Mar 2, 2025Coding GRPO from Scratch: A Guide to Distributed Implementation with QWen2.5–1.5B-InstructGRPO excel in math and logic tasks by learning to generate a CoT. this detailed overview of how R1 and R1-Zero were trained.
InTowards DevbyArXiv In-depth Analysis·Nov 3Adding Agents Made It Worse, Until the Orchestrator Learned to Write ProgramsWhy adding agents degrades an untrained orchestrator, how programmatic workflow generation fixes the coordination tax, and what 69…A response icon1A response icon1
ArXiv In-depth Analysis·Oct 29Two Halves of One Brain: What RynnUnify Learns When Prediction and Understanding Share WeightsA 16B model that plans in video, then turns its own imagined futures into joint torques — and the places where that loop breaks.
InGoPenAIbyArXiv In-depth Analysis·Oct 29Frontier Models Look Excellent Until a Human Has to Do What They SayThe benchmark figures on technical report slides routinely show frontier multimodal models posting strong scores on visual question…
InTowards DevbyArXiv In-depth Analysis·Oct 27The 5% Compliance Blindspot: What Peking University and Ant International Learned Testing AI Agents…Why frontier LLMs with strong end-goal performance struggle in real enterprise operations: irreversible ledger mutations, phantom…A response icon1A response icon1
InGoPenAIbyArXiv In-depth Analysis·Oct 26The 25-Second Deliberation Gap: Why Frontier Multimodal Models Still Struggle as Autonomous…The Joint Research Team evaluated GPT-6-Astra across 12 benchmarks and 8 physical robot platforms. Here is the unvarnished systems…A response icon1A response icon1