technology
The class of AI models, such as GPT, Gemini, and Claude, that generate text by predicting the next word. In episode 01 they are both Syntopic's workhorse and the reason it needs checks: fluent, fast, and confidently wrong. Start at about 29 minutes.
Across the episodes
01
Enters as the double-edged engine: the labor that makes the project affordable and the source of the errors it has to catch
What Chad saw at the conferences: LLM-assisted knowledge graphs built from written texts, with no one applying the method to recorded speech.
"They're using an LLM to read the text and build a knowledge graph,"Chad Nantais, Syntopic, Episode 1 (11 Oct 2026), 22:23
The premise of Syntopic's gates: because generation is statistical, an LLM's output is a candidate to be checked, never a record.
"LLMs are probabilistic. they're using statistics to guess what the next word should be based on the data that it has about the past."Chad Nantais, Syntopic, Episode 1 (11 Oct 2026), 29:54
Connections
checked byHuman in the Loop
Deterministic code, then an LLM, then a human.