Grok 4 is literally #1 on every single major benchmark. Read our blog: x.ai/news/grok-4

Becoming an RL diehard in the past year and thinking about RL for most of my waking hours inadvertently taught me an important lesson about how to live my own life. One of the big concepts in RL is that you always want to be “on-policy”: instead of mimicking other people’s

"Deep Researcher with Test-Time Diffusion" This paper treats report writing as an iterative retrieval‑augmented diffusion process that can be enhanced by component‑wise self‑evolution. This demonstrates SoTA on multi‑hop search‑and‑reasoning benchmarks.

Apple just released Embedding Atlas: An open-source visualization tool for your embeddings. I just gave it a quick spin with some data stored in my vector database. These are my first impressions: - Nice exploration UX with hover and tool tip for single data points - Shows you

[0/3] 🚀 Introducing Verlog – an open-source RL framework built specifically for training long-horizon, multi-turn LLM agents. 📊 Max episode length comparison: •VeRL / RAGEN → ~10 turns •verl-agent → ~50 turns •Verlog (ours) → 400+ turns 🔥 ⚙️ Technical foundation:

Summary of the findings for the Hierarchical Reasoning Model on ARC

Keynotes from RLC 2025 are out! youtube.com/playlist?list=…

BRILLIANT Google DeepMind research. Even the best embeddings cannot represent all possible query-document combinations, which means some answers are mathematically impossible to recover. Reveals a sharp truth, embedding models can only capture so many pairings, and beyond that,

Let’s talk about VLAs in robotics 🤖 (Vision-Language-Action models) A relatively new type of robotics policies that bring the power of LLMs into the real world. If you’ve seen robots folding laundry, washing dishes, or cleaning rooms – chances are they used something VLA-like.

RLAD (Reinforcement Learning with Abstraction and Deduction) trains models via RL using a 2-player setup: ▪️ An abstraction generator – proposes short, natural-language “reasoning hints” (abstractions) summarizing key facts and strategies. ▪️ A solution generator – uses them to

🚀 Exciting updates in Qwen Code v0.0.12–v0.0.14! ✨ What’s new? • Plan Mode: AI proposes a full implementation plan—you approve before a single line changes. • Vision Intelligence: Auto-switch to vision models (Qwen3-VL-Plus with 256K input / 32K output!) when images

manage Jira issues from your terminal with a text-based interface

Your Base Model is Smarter Than You Think This paper proposes a way to beat the lack of generation diversity in RL without RL! By using Markov Chain Monte Carlo’s ‘power sampling’ that reuses a base LLM’s own probabilities, it’s able to beat GRPO without training & verifiers

AI agents are supposed to collaborate with us to solve real-world problems, but can they really? Even the most advanced models can still give us frustrating moments when working with them deeply. We argue that real-world deployment requires more than productivity (e.g., task

🔥 GPT-6 may not just be smarter. It literally might be alive (in the computational sense). A new research paper, SEAL: Self-Adapting Language Models (arXiv:2506.10943), describes how an AI can continuously learn after deployment, evolving its own internal representations

Gemini be like

We keep scaling model parameters by increasing width and stacking more layers, but what if the truly missing axes for continual learning are compression and stacking the learning process? Excited to share the full version of Nested Learning, a new paradigm for continual learning

Hugging Face has released a 214-page MASTERCLASS on how to train LLMs > it’s called The Smol Training Playbook > and if want to learn how to train LLMs, > this GIFT is for you > this training bible walks you through the ENTIRE pipeline > covers every concept that matters from

Mixture-of-Experts (MoE) is a powerful way to scale large language models (LLMs): instead of running the full model for every token, a router activates only a few “experts,” giving more capacity at roughly the same compute. But routing is still a sore spot. Most MoE systems use