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.

UNREAL: Unifying Retrieval and Long-Context with a Single Model

UNREAL is a model-native evidence selection framework that unifies corpus retrieval and long-context inference by adding minimal trainable parameters to frozen LLMs. It significantly improves recall on retrieval benchmarks and accuracy on long-context tasks while reducing computational costs.

arxiv.org →

Connections

Supports

Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning UNREAL's model-native approach to deciding what evidence to retrieve parallels meta-reasoning about when and how to allocate inference - both concern intelligent pre-processing before expensive reasoning steps
Unreal Agent: Cost-Efficient AI Agent Harness UNREAL's evidence selection framework reduces unnecessary context passed to LLMs, directly enabling the kind of cost-efficient agent harness design that Unreal Agent targets - better retrieval precision means fewer wasted tokens

Related to

Context Language Models Context Language Models directly address how LLMs handle and process context windows - UNREAL's long-context inference component addresses the same architectural concern of maximizing what models can do with available context
Semantic ID-based Generative Retrieval for Podcast Discovery at Spotify Both address retrieval from large corpora using learned representations - UNREAL unifies retrieval with LLM inference while Spotify's work uses semantic IDs for generative retrieval in recommendation contexts
Contrastive Language Model (CLM) Contrastive Language Models use contrastive objectives to improve retrieval - UNREAL similarly aims to bridge the gap between retrieval and generation in a unified model, addressing the same train/inference objective mismatch

Challenged by

How Long Contexts Fail UNREAL model aims to unify retrieval and long-context processing, which directly addresses the failure modes described - good retrieval could mitigate context poisoning and distraction by being selective about what enters the context

Bag of Decisions Reranker Both address the retrieval quality problem — UNREAL unifies retrieval and long-context in one model, while Bag of Decisions improves reranking post-retrieval via multi-question rubrics; complementary approaches to the same search relevance challenge

Type:	link
First published:	Thursday, 8 October 2026 at 19:53 UTC
Location:	https://arxiv.org/abs/2610.08463v1
URL:	https://mattwood.fyi/i/f2753f78

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