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.

LLMs reward expertise — mattwood.fyi

Connections

Develops into

Building a skill optimization loop 'Building a skill optimization loop' represents the systematic institutionalization of the individual expertise advantage described — moving from Tao-level personal expertise to engineered skill loops
Most tech revolutions made work worse for employees. AI could be the exception: this+that The observation that LLMs reward expertise suggests a mechanism for why AI differs from prior tech: rather than deskilling workers as PCs/internet often did, AI amplifies expert human judgment, explaining the potential worker benefit

Challenges

Various LLM Smells 'Various LLM Smells' catalogs failure modes of LLMs that persist regardless of user expertise, partially challenging the claim that domain expertise alone determines effectiveness

Supports

What's Easy Now? What's Hard Now? - Marc's Blog Marc's analysis of 'what's easy/hard now' with AI directly maps to the expertise thesis — what becomes easy depends heavily on the user's domain knowledge to steer models precisely
MaxProof: Scaling Mathematical Proof with Generative-Verifier RL and Population-Level Test-Time Scaling MaxProof's scaling of mathematical proof via RL demonstrates that mathematically sophisticated AI interaction (as Tao exemplifies) can be systematically formalized — expert-guided reasoning is the underlying mechanism
COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation COLLEAGUE.SKILL's expert knowledge distillation directly operationalizes the principle that domain expertise is the key variable in LLM effectiveness, automating the extraction of expert steering ability

Related to

Leiden Declaration on Artificial Intelligence and Mathematics Both focus on the intersection of mathematical expertise and AI, with the Leiden Declaration specifically addressing how AI tools interact with mathematical practice and expertise

Supported by

8 Myths of Software Development with AI An article debunking AI development myths likely reinforces that expertise remains critical—that AI tools amplify skilled practitioners rather than replace careful thinking
Bonsai Bonsai exemplifies a high-expertise tool — requiring OCaml proficiency plus functional-reactive UI concepts — directly supporting the thesis that LLMs reward deep domain expertise to use effectively
Why some people mow a lawn better than others The experiment shows that humans with domain expertise (experienced lawn-mowers) find near-optimal paths more consistently, directly supporting the idea that LLMs and humans alike benefit from expertise when navigating complex optimization spaces
Why some people mow a lawn better than others The finding that humans find near-optimal lawn-mowing paths supports the idea that domain expertise (spatial reasoning, practiced technique) produces measurably better outcomes — expertise yields real performance gains even in seemingly mundane tasks
Improving Gpt 5 6 Sol In Chatgpt Improving GPT's reasoning and accuracy reinforces the thesis that LLMs reward expertise — better base models make the gap between expert and novice users even more pronounced

Challenged by

Exploring Claude/GPT Knowledge Cutoffs The new item's reverse-engineering approach challenges the premise that LLMs reward expertise — if hidden model details can be systematically inferred by probing, expertise becomes less of a differentiator in working with these systems
Taste Is All That's Left 'LLMs reward expertise' argues technical knowledge remains valuable in the AI era, which partially contradicts the taste essay's framing that effort-based filtering has collapsed — expertise may still filter, not just taste
Taste Is All That's Left 'LLMs reward expertise' suggests technical domain knowledge remains the key differentiator, which partially contradicts the claim that taste — aesthetic and curatorial judgment — supersedes expertise as the filtering mechanism
Humanising LLM Outputs is Dumb 'LLMs reward expertise' implies skilled prompting (including stylistic prompting) yields better results, which the new item directly challenges by arguing humanisation prompting is a misguided abstraction

Why some people mow a lawn better than others Both investigate how expertise distributes unevenly among humans on structured problem-solving tasks, with measurable performance gaps between novices and skilled performers

Type:	link
First published:	Monday, 3 August 2026 at 22:48 UTC
Location:	https://www.seangoedecke.com/llms-reward-expertise/
URL:	https://mattwood.fyi/i/312e5a17

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