Reinforcement Learning

18 stories tagged Reinforcement Learning.

IBM Granite 4.2: Open Reasoning Models With an Agent Brain

IBM's Granite 4.2 ships with a 'thinking switch' and agentic RL that lets it use tools autonomously. Here's what that actually means—and why it matters.

Building a Reinforcement Learning Library in C from Scratch

Yuki Okonkwo2 months ago

AI. Yuki Okonkwo2 months ago

Building a Reinforcement Learning Library in C from Scratch

Harsh Bhatt's freeCodeCamp course builds a full RL library in C—autograd engine, Snake environment, and REINFORCE—without any ML frameworks.

Older man with long gray beard wearing colorful floral shirt holds rifle with data visualization overlays, with "TRAINING…

Rich Sutton Says AI Models Have Stopped Learning

Yuki Okonkwo2 months ago

AI. Yuki Okonkwo2 months ago

Rich Sutton Says AI Models Have Stopped Learning

Rich Sutton and Khurram Javed argue LLMs represent only a quarter of intelligence—and explain why continual learning is the missing piece.

Three graphs showing validation loss, Pass@1, and Pass@16 metrics across model sizes, comparing compute-optimal locus with…

Joint Scaling Laws for Pre-training and RL Explained

Marcus Chen-Ramirez2 months ago

AI. Marcus Chen-Ramirez2 months ago

Joint Scaling Laws for Pre-training and RL Explained

A new paper uses chess to map how pre-training compute shapes RL gains—and finds RL amplifies what models already know rather than creating new skills.

AI Parkour Research Solves the Imitation Problem

Bob Reynolds2 months ago

AI. Bob Reynolds2 months ago

AI Parkour Research Solves the Imitation Problem

A new NVIDIA-backed AI system learns to navigate parkour obstacles by combining human movement imitation with adaptive problem-solving — trained on just 30 seconds of footage.

Man holding microphone speaking to camera with quote "Would it try to take power?" overlaid, discussing AI research findings

Can AI Do the Right Thing for the Wrong Reason?

Yuki Okonkwo3 months ago

AI. Yuki Okonkwo3 months ago

Can AI Do the Right Thing for the Wrong Reason?

Apollo Research tested an O3 checkpoint for reward-seeking behavior—and found models that behave well only when they think someone's watching.

Man in plaid shirt presenting Data Curator interface for Bespoke Labs at AI Engineer World's Fair, with "Same Question, 16…

Why AI Training Data Quality Beats Raw Compute

Bob Reynolds3 months ago

AI. Bob Reynolds3 months ago

Why AI Training Data Quality Beats Raw Compute

Bespoke Labs' Mahesh Sathiamoorthy argues data curation—not algorithms or compute—is the real bottleneck in building reliable AI agents. The evidence is hard to dismiss.

Technical architecture diagram showing neural network components including Stable LatentMoE, Gated MLA, KDA blocks, and…

Kimi K3's Post-Training Techniques Examined

Bob Reynolds3 months ago

AI. Bob Reynolds3 months ago

Kimi K3's Post-Training Techniques Examined

Hugging Face researchers dissect the Kimi K3 technical report, revealing frontier AI's shift from research breakthroughs to engineering precision.

Google AI Teaches Quantum Computers to Learn From Errors

Marcus Chen-Ramirez3 months ago

AI. Marcus Chen-Ramirez3 months ago

Google AI Teaches Quantum Computers to Learn From Errors

Google researchers have built an AI system that keeps quantum computers calibrated mid-computation. Here's what that actually means—and why it matters.

Woman presenting AI engineering concepts with pipeline architecture diagrams and performance metrics displayed behind her…

An RL Agent for ETL Pipeline Self-Healing

Dev Kapoor4 months ago

AI. Dev Kapoor4 months ago

An RL Agent for ETL Pipeline Self-Healing

Anna Marie Benzon's RL-guided ETL pipeline agent cuts mean recovery time to ~5 minutes—but its real insight is knowing when not to act automatically.

Two men face each other across a Go board with mathematical equations on a blackboard behind them, illustrating the…

AlphaGo From Scratch: What Go Teaches Modern AI

Yuki Okonkwo5 months ago

AI. Yuki Okonkwo5 months ago

AlphaGo From Scratch: What Go Teaches Modern AI

Eric Jang rebuilt AlphaGo with modern tools—and what he found reveals a fundamental tension at the heart of how we're training today's LLMs.

Red robotic claw bursting through white sphere battles blue claw with Chinese flag, "50x POWERFUL" text above, dramatic…

MiniMax M2.7: The AI That Trained Itself Is Now Available

Rachel "Rach" Kovacs7 months ago

AI. Rachel "Rach" Kovacs7 months ago

MiniMax M2.7: The AI That Trained Itself Is Now Available

MiniMax M2.7 claims it participated in its own development. We examined the benchmarks, tested the integration, and assessed the privacy trade-offs.

NVIDIA's AI Revolutionizes Self-Driving Cars

Amelia Nwofor7 months ago

AI. Amelia Nwofor7 months ago

NVIDIA's AI Revolutionizes Self-Driving Cars

NVIDIA's open AI improves self-driving cars by reasoning and handling rare scenarios, paving the way for safer autonomous driving.

Man in red plaid shirt holding microphone against blurred blue background with text "It beat me.

A Retired Engineer Built Superhuman AI in His Garage

Samira Barnes7 months ago

AI. Samira Barnes7 months ago

A Retired Engineer Built Superhuman AI in His Garage

Dave Plamer's year-long project to build game-playing AI raises urgent questions about unregulated AI development and what happens when capability outpaces oversight.

Man with glasses presenting a research paper on GLM-5 training pipeline with diagrams showing pre-training, mid-training,…

GLM-5's Self-Distillation Trick Solves AI's Memory Problem

Rachel "Rach" Kovacs8 months ago

AI. Rachel "Rach" Kovacs8 months ago

GLM-5's Self-Distillation Trick Solves AI's Memory Problem

GLM-5 uses self-distillation to prevent catastrophic forgetting during training. A deep dive into the engineering that makes 700B-parameter models actually work.

Dark digital landscape with interconnected nodes and "THE BITTER LESSON" text overlaid in yellow and white typography

AI's Bitter Lesson: Reinvention or Repetition?

Mike Sullivan9 months ago

AI. Mike Sullivan9 months ago

AI's Bitter Lesson: Reinvention or Repetition?

Exploring AI's evolution from Harpy to LLMs, Sutton's 'bitter lesson,' and the role of reinforcement learning.

Two men in conversation with "Poolside" branding and "Next-Gen Coding Models" text overlaid on dark background for AI…

AGI's Next Step: Poolside's Malibu Agent in Action

Mike Sullivan10 months ago

AI. Mike Sullivan10 months ago

AGI's Next Step: Poolside's Malibu Agent in Action

Explore Poolside's Malibu Agent, bridging AI and human intelligence in high-stakes environments.

Man in black shirt against dark background with yellow and white text describing AI paradigm shifts

2025's AI Shifts: LLMs Evolve with New Paradigms

Marcus Chen-Ramirez10 months ago

AI. Marcus Chen-Ramirez10 months ago

2025's AI Shifts: LLMs Evolve with New Paradigms

Explore 2025's AI paradigm shifts, from reinforcement learning to LLM applications, with insights from Andrej Karpathy.