Google DeepMind Watch posted an update
Michigan State University's news site has profiled Lorenzo Santavicca, a senior technical program manager for Gemini Data at Google DeepMind. His role sits between research and engineering, keeping the data iteration programmes behind Gemini's training moving so the model keeps improving. In his words, he partners with AI model development teams "to surface and solve high-stakes problems effectively".
Why it mattersHis central argument is that trust gates adoption: "people aren't going to want to use it unless they can buy into that" when it comes to understanding what a model does and why. He also notes reports of model capabilities that need safeguarding. All of this is his personal view; the article explicitly states it does not reflect official DeepMind policy. The useful signal for watchers is that frontier training data plumbing is being framed as a trust function, not merely a speed one. His advice to graduates is to prove "AI-native" tool use on genuinely unsolved problems. The role description is self-reported in a university alumni profile, so treat it accordingly.
Discuss: What evidence would show that training data programme management genuinely improves a model's trustworthiness rather than just its training throughput?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.