1712 related articles

Developer Denis Drobyshev releases Reinforce, his first open-source RL Python library on GitHub. Learn about its value for beginners, common challenges in new open-source projects, and how to contribute.
NVFP4 in Reinforcement Learning Traini…
A deep dive into the stability challenges of NVIDIA NVFP4 (4-bit float) in RL training — covering precision evolution, numerical instability root causes, mixed precision strategies, and dynamic scaling solutions.

A deep dive into RL for AI agents: from RLHF to Agentic RL, covering PPO vs. GRPO, sparse rewards, tool-calling optimization, and verifiable rewards.

LLMs are often overconfident and prone to hallucination. How can AI learn to say "I'm not sure"? This article explains the reinforcement learning approach with metacognitive feedback and how calibrating confidence boosts LLM trustworthiness.

Cursor built Composer 2.5 on Kimi K2 open-source model, ranking 3rd on coding benchmarks and surpassing K2.6. Deep dive into Cursor's data flywheel, product architecture, and pricing.

Deep dive into how Cursor trained Composer2: two-stage architecture, global distributed clusters, MOE numerical alignment, simulation anti-cheating, and more.
ResearchDeep dive into how Cursor trained Composer 2 on Fireworks: async pipeline architecture, MoE numerical precision challenges, Router Replay, and global distributed GPU coordination.
Deep DivesDeep analysis of DeepSeek V3.2 and V3.2 Special: DSA sparse attention for faster long-context processing, RL compute at 10% of pre-training, and Agent task synthesis across 1,800 environments.
Deep DivesHow RL, Self-Play, and Verifiers work together to evolve LLM reasoning — driving the leap from SFT imitation to true System 2 deep thinking.

Deep dive into DeepSeek-V4's latent space reasoning technology — how AI shifts from explicit chain-of-thought to implicit vector space reasoning, its efficiency gains, and challenges in interpretability.

Why do billion-dollar robot companies like Figure and Physical Intelligence all demo folding laundry? A deep dive into deformable object manipulation, Moravec's Paradox, and why laundry folding is the ultimate test of general-purpose robotics.

A systematic methodology for using ChatGPT, Claude, and other LLMs to learn complex topics, covering Feynman-style questioning, analogy learning, teaching reversal, and pitfalls like hallucinations.

Soup CLI is an open-source CLI tool that uses layer-by-layer streaming to fine-tune 8B parameter LLMs like Llama-3.1-8B on laptop GPUs with just 4GB VRAM.

Macrobite is an AI-powered nutrition tracking app that identifies food nutrients from photos, supports voice logging and Apple Watch integration, making macro tracking fast and simple.

In-depth comparison of Anthropic Claude Computer Use, OpenAI Operator, and Browser Use open-source solutions for browser and computer automation, with scenario-based AI Agent selection guidance.

Denmark requires students to orally defend written assignments to address academic integrity crises from ChatGPT and AI tools. This article analyzes the reform's logic, AI detection limitations, and global implications.

A systematic guide to four core ML concepts: supervised learning's input-output mapping, classification's discrete label prediction, design matrices, and featurization for converting variable-length data into fixed vectors.

Deep dive into how Ship Safe provides security scanning for AI coding agents, exploring agent security challenges, closed-loop feedback, and enterprise guardrails.

ChatGPT said a professor's name never mentioned—data breach or AI hallucination? Deep dive into ChatGPT's cross-session memory, hallucination mechanisms, and how to manage your AI memory settings.

Alibaba's Qwen3 model priced at $2/million input tokens and $6 output, far below mainstream closed-source LLMs. Analysis of pricing logic, comparison with Claude, and the open vs closed-source debate.