Julian Schröter

@julianschroeter.bsky.social

Munich, LMU | Computational Literary Studies (ML + literary history and theory)
https://julianschroeter.github.io/
https://www.germanistik.lmu.de/schroeter

@mariaa.bsky.socialReposted by @julianschroeter.bsky.social

It's infuriating how the U.S. is treating our international researchers. We are so incredibly lucky that talented people want to come to our country to study, research, work, and build. Trump, Musk, and their cronies are destroying the best parts of science in the U.S.

@tedunderwood.comReposted by @julianschroeter.bsky.social

This is quite likely to have a bad time in the courts. They have looked at similarly large "fees" and blinked and said "you are aware that taxation power belongs to Congress, right?"

@tedunderwood.comReposted by @julianschroeter.bsky.social

Models like Talkie-1930 sound like voices from the past. If they could reliably speak from specified historical vantage points, researchers might also use them to simulate the past. But how reliable are they? Today we release a benchmark answering that question for English contexts 1831-1930.

Free-text evaluation of answers to character modeling and constrained generation questions. This is just one of several scores Chronologic-EN can produce; we focus on it here because it's both the hardest test and the one most relevant to simulation of the past. Frontier models reach 72%; Talkie-1930 is stronger than several larger competitors, but not at the frontier by this measure. Note that this score has improved ~45% in the last two years, but still falls perceptibly short of ground truth.