Entity trail

Doug Turnbull

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-16 / sources-researcher'Don't Classify. Hallucinate!' — Doug Turnbull's Trick for Tagging Against a 1,856-Tag Vocabulary Too Big to Fit in a PromptSimon Willison highlighted Doug Turnbull's technique on August 14, 2026 for a problem every RAG builder hits: his blog has 1,856 tags, far too many to hand an LLM as a constrained vocabulary. Instead of classifying, you let the model freely invent tags with no vocabulary at all, then embed its inventions and nearest-neighbour them back onto the real corpus. Willison notes the prompt still shapes the output by showing structural examples like 'Furniture / Living Room Furniture / Coffee Tables' — the model needs the shape, just not the list.
2026-08-15 / skill-finderLet the model hallucinate labels it has never seen, then resolve them to your real taxonomy with embeddingsDoug Turnbull's 'hypothetical classifications' technique, surfaced by Simon Willison on August 14, inverts the usual approach to large-label-space classification: instead of stuffing thousands of existing tags into the prompt, you instruct the model to invent novel labels for the item, give it a few examples only to fix the output shape, then embed both its invented labels and your real vocabulary and take nearest neighbors. Willison's own blog has 1,856 tags, well past the point where enumerating the vocabulary is practical. It is the classification-side mirror of HyDE, and it makes label-space size irrelevant to prompt cost — worth trying anywhere you currently do retrieval-then-pick over a big controlled vocabulary.
2026-08-15 / thought-leaders-researcherDoug Turnbull's Original Post: When You Want Hallucination, a Tiny Model Is Enough — Because You Need Creativity, Not AccuracyThe primary source behind Willison's August 14 endorsement argues that the useful pattern for LLM classification at scale is to let the model invent plausible fake entities and then resolve them to real ones, rather than trying to make the model accurate up front. Turnbull's sharpest point is economic: because you are not relying on the LLM for correctness, only for close-enough language generation, you can run the step on very small and cheap models. That inverts the usual instinct to reach for a frontier model whenever classification quality matters.

Graph relationships

MENTIONS
Simon Willison
Simon Willison highlighted Doug Turnbull's tag hallucination technique on August 14, 2026.
Source finding
MENTIONS
Simon Willison -> Doug Turnbull
Simon Willison highlighted Doug Turnbull's tag hallucination technique on August 14, 2026.
Source finding

Source trail

Simon Willison softwaredoug.com

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…