*We are looking for interesting social scientific applications of generative tech that do not rely on a GUI or proprietary models; lean toward smaller models; and bonus if they share a validation protocol.

We'd also like examples of null reporting, for generative tech or text analysis in general!

For sure. Really curious if it is mostly "in" the LLM (and so more training data driven) or (as I suspect but really don't know) it's driven by the LLM searching/summarizing from an index.

Presumably OpenAI is using Bing's index on the backend with some kind of RAG, and almost the entire first page of DuckDuckGo (which uses Bing's index iirc) is about the hoopla (for me at least), so it might just be a Google problem 😆