Super excited about this - the first results from this were INSANE compared to gpt-4o for our use case.
Question though - and it could be implementation - but if I upload and do the first run with o3-mini I get nonsense like this:
In other words, while the screenplay shows various narrative tensions—such as internal struggles, interpersonal conflicts, and evolving relationships—it does not overtly break its conflict into the four distinct categories defined by the model. The text offers compelling dramatic moments (as seen in the intense dialogue between Dane, Genevieve, Mia, Natalie, and others citeturn0file0 citeturn3file3Reviewing “Beside Ourselves – Gabriel Mizrahi.pdf” in detail, I see a richly layered narrative with multiple interwoven character arcs, conflicts, and emotional shifts. However, when I look for clear, formal divisions echoing the Universal Narrative Model’s “four throughlines” (that is, the Objective Story Throughline, the Main Character Throughline, t
but then the run gets marked failed and dies.
Yet, if I do the first run with gpt-4o, and then switch over to o3-mini, the results are crazy good and it works from then on as expected.
(sry, for context, my approach is make vectorstore, attach vectorstore to thread, then run)
