Meditation

Keeping a Vision Pipeline Honest: Three Hallucination Guards I Built for a Video-Captioning Agent

How Dragon Commentary Studio separates perception from schema, filters OCR by consensus, and gates every caption through a critic loop.

I Trained a Board-Game AI From Zero Using Self-Play (and Broke It Several Times First)

How Caro5's bot went from a frozen laptop and a useless first model to a generate → train → arena → promote loop running on Modal.

RLHF and Beyond: How Reinforcement Learning Taught Language Models to Behave

Reinforcement Learning · Part 4

A base language model fresh out of pretraining is a strange creature. It has read a large fraction of the internet and can continue any text with uncanny fluency — but it isn't trying to help you.

Policy Gradients and PPO: Learning Behavior Directly

Reinforcement Learning · Part 3

DQN taught me one way to act intelligently: learn the value of every action, then pick the best one. But there's a second lineage in reinforcement learning with the opposite philosophy — skip the values, and optimize the behavior itself.

DQN: The Moment Reinforcement Learning Met Deep Learning

Reinforcement Learning · Part 2

In 2013, a small London startup called DeepMind posted a paper showing a single algorithm learning to play Atari games — from raw pixels, with no game-specific knowledge, using only the score as feedback.

Reinforcement Learning from Zero: Agents, Rewards, and the Loop That Learns from Consequences

Reinforcement Learning · Part 1

Supervised learning felt intuitive to me from day one: here's the input, here's the right answer, minimize the difference.

Below Four Bits: QAT, BitNet, and the Race Toward One-Bit Intelligence

Quantization · Part 4

Everything in this series so far has been post-training quantization: take a finished model, compress it, hope the damage is small.

GGUF and the Art of Choosing a Quant: A Field Guide to Local Models

Quantization · Part 3

Open any popular model's page on Hugging Face and you'll find the GGUF listings: Q4KM, Q5KS, Q6K, Q80, IQ2XS — a wall of cryptic suffixes, each a different point on a size-quality curve, downloaded millions of times by people running models on gaming PCs an…

The Outlier Problem: The Discovery That Shaped Modern Quantization

Quantization · Part 2

Here's a puzzle that stumped the field around 2022. Quantization recipes that worked beautifully on small language models — clean INT8, minimal quality loss — fell off a cliff on big ones.