What is mattwood.fyi?

A live, structured feed of riffs and links from Matt Wood, Chief AI & Technology Officer at AWS — readable by people and queryable by agents as a typed knowledge graph.

Is there an API and is it free?

Yes. A public, read-only query API is available with no authentication and no API keys. See https://mattwood.fyi/openapi.json and https://mattwood.fyi/docs.

How should an AI agent use this site?

Fetch https://mattwood.fyi/feed.json for the full structured feed, or query the API for keyword/semantic search and the connection graph. See https://mattwood.fyi/agents/ for when-to-use guidance.

Engineering a Calibrated Classifier Model

Marc Brooker describes building Hobson, a 2-billion parameter classifier optimized for accuracy and calibration at low latency, by fine-tuning a pre-trained LLM with a pointer head instead of a language modeling head to score multiple choice answers.

brooker.co.za →

Connections

Supports

Kev: Tiny Decision Models on Qwen Both demonstrate the trend of building small, task-specific decision/classification models from LLMs - Kev on Qwen and Hobson both prioritize efficiency and calibration over raw scale
The Plunging Price of Thought Hobson's design - optimizing a 2B parameter model for low latency classification - is a practical engineering response to inference cost pressures described in 'The Plunging Price of Thought'

Related to

Evaluating LLM Judge Agreement and Reliability Both deal with reliability and calibration of LLM-based judgment systems - Hobson explicitly optimizes for calibration while this item evaluates LLM judge agreement and reliability
LLM Classification Is Feature Engineering Both address using LLMs for classification tasks - Hobson uses a fine-tuned LLM with pointer head, while 'LLM Classification Is Feature Engineering' examines the paradigm of LLM-based classification from an engineering perspective
Contrastive Language Model (CLM) Both involve architectural modifications to standard LLMs for classification purposes - CLM uses contrastive objectives while Hobson replaces the language head with a pointer head to optimize for classification calibration
Training a 4B Model for Faster SQL Query Plans Both describe fine-tuning smaller/specialized models (4B SQL model vs 2B Hobson classifier) optimized for specific tasks with latency/performance tradeoffs rather than using frontier models

Type:	link
First published:	Monday, 28 September 2026 at 18:12 UTC
Location:	https://brooker.co.za/blog/2026/09/28/engineering-system-one.html
URL:	https://mattwood.fyi/i/368d6d67

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

Connections