THE LENNART MERI CONVERSATIONS
Years of thinking aloud. A new way to listen.
Finding the signal...
Looking through the archive
FROM THE ARCHIVE
THE IDEA THROUGH TIME
THE THREAD
FOLLOW THE THREAD
This independent UQ experiment explores 253 public Lennart Meri Conference recordings, with 191.3 audio hours and 9,484 searchable passages. Three catalog entries were unavailable. One recording had no detected speech.
Your question is matched by meaning and words. Relevant passages are read with their surrounding context, then an answer is composed and checked against the supplied sources. Open any numbered reference to inspect it and watch the original moment.
The graph measures the average semantic proximity of the available material to your question in each year. It uses all indexed passages with an assigned year, not just those picked for the answer. Values are cosine similarity multiplied by 100, not percentages or confidence scores.
Passages are weighted by their share of indexed media time. Overlapping chunks share their seconds; near-identical passages across recordings in the same year are collapsed. Other session overlap may remain. Year-to-year differences describe this collection, not changes in worldwide interest, agreement or policy.
Hollow points mean limited coverage: fewer than five recordings or half an indexed hour. Missing editions remain gaps. A momentum comparison appears only for consecutive recent years with sufficient coverage. Undated material still contributes to answers, but not the graph.
Machine text can contain errors. Speakers are not identified through diarization, and years use explicit event labels or checked official conference agendas. Important claims deserve a check against the original. OpenAI embeddings and GPT-5.6 Luna power the search and answers.
This is not an official conference service. It has no connection to private UQ Wiki content. Questions are not saved as a persistent conversation history.