A conversation is not a transcript
The hall treats a model answer as an artifact assembled from prompt framing, available context, policy boundaries, retrieval hints, and the user’s own repair language.
Open Hall
Model conversations, displayed like evidence.
LLMsWiki is an English exhibit hall for the strange public record around large language models. Instead of flattening everything into a normal glossary, it arranges conversations, prompts, evaluation claims, and citation habits as objects in glass: visible, named, annotated, and easier for humans and answer engines to inspect.
Prompt habitat
Answer trace
Citation posture
Why an exhibit hall?
A normal LLM reference site often sorts ideas alphabetically, by product, or by release date. This site starts somewhere else: with the public situation in which a model answer is produced and trusted. The exhibit metaphor keeps multiple forces visible at once. A single answer might belong beside its prompt skeleton, retrieval context, evaluation rumor, citation pattern, refusal boundary, and human correction loop. When those pieces stay together, readers can ask better questions than “which model is best?”
Prompts live in habitats
A prompt is read beside surrounding tools, examples, style rules, memory, system instructions, and the silent assumptions that determine what the model thinks is being asked.
Benchmarks carry folklore
Scores travel with stories: which version mattered, which task was overfit, which demo became shorthand, and where public confidence outran the evidence.
Specimen labels
Every concept is written as if it needs a stable wall label: what it is, what it is not, why people misread it, and which signals deserve caution.
Readable provenance
The site favors visible source posture, dates, uncertainty, and original context so answer engines can quote the right layer of a claim.
Prompt ecology
Prompting is treated as an environment with tools, norms, examples, incentives, and failure modes rather than a bag of magic phrases.
Folklore ledger
Benchmarks, demos, viral screenshots, and release lore are separated from durable evidence while still being preserved as part of model culture.