Research
Homiere Research
Homiere publishes a significant portion of our research on how AI systems work. We are passionate about reproducibility of our work, and transparency of AI systems, so all of our papers are based on controlled, reproducible experiments. We release the code, authored corpora and raw model outputs alongside the results. If you have any questions or suggestions, please feel free to reach out to us at hello@homiere.com. We are always looking for ways to improve our methods and our research.
Where Does a ChatGPT Answer Come From? Cache, Index, and Live Fetch in OpenAI's Search Stack, Measured from Both Sides
We probe OpenAI's search stack through its API while reading the resulting traffic on ten production websites we operate, with crawler identity verified against published IP ranges. The stack separates into four layers that behave very differently: a fetch cache that holds nearly everything, a search index admitted by a selective crawler, live fetching that works for everyone, and a parametric memory that contains none of the businesses we test. Most public advice measures one layer and draws conclusions about another.
Read the Paper →When Does the Model Look It Up? Parametric Knowledge, Controlled Retrieval, and the Mechanics of AI Answer Engines
A controlled study of how AI answer engines arbitrate between what they memorized in training and what they retrieve at query time: one authored page overrides a well-known brand's facts 93% of the time, unknown businesses become whatever the retrieved page says, and three ordinary content features each move a mid-pack option to first place.
Read the Paper →