23 October 2026 - Rapperswil Back in Rapperswil for day 2. The day started off with Pavel Sulimov talking about open source contributions. He found that he needed to search for a better tool rather than open a PR on an unmaintained one and built a tool, consta, to answer that question. Then Hilal Isak gave a rather polemic talk about missing female data which presented running data on a sample size of one runner. Vadim Vlasov and Eric Glaser next talked about an as-yet-unreleased game to teach AI. I was at a company that started out on that project before pivoting a couple startups back, so I do not judge their prospects highly but they seem to be having fun. After that Vince Nelidov talked about data credibility with lots of Bayesian statistics on a toy dataset. The last talk before lunch was Olena Kutseno's on data privacy. She gave a good look at data pipelines and gave good arguments for left-shifting most PII. But she also gave a number of examples where seemingly non-personal data could be traced back to a person. After lunch Dominik Treaxl gave a talk on a tabular prior-fitted network. Then Emmanuel Okedele gave an insightful prespective on evaluating open-weight LLMs. Since I'm using those a good bit these days it was quite useful. The last scheduled speaker of the day was Almudena Carrillo, who spoke about auditing Instagram data. Again it was a sample size of one though. And then the day ended again with lightening talks. |
Last changed on 2 November 2026 by Bradley James Wogsland.
Copyright © 2026 Bradley James Wogsland. All rights reserved.