Donald Clark has already gotten a couple posts out of Google's EdAI announcements. This paper (86 page PDF) presents Google's efforts to "translate high level principles from learning science into a pragmatic set of seven diverse educational benchmarks" and to "develop a new set of fine-tuning datasets to improve the pedagogical capabilities of Gemini." To develop, in other words, a generative AI tutor. It deserves a careful reading, and your impressions of the 35 page main text and 50 pages of supplementary material may vary from mine. While the paper is very up-to-date with respect to AI, it reveals (to me) a dated and psychology-heavy understanding of 'learning science' that seems limited to work on intelligent tutoring systems - but you can view their 'Pedagogy rubrics' in section 4.3.1. Saying, "the gen AI models that power most of the latest EdTech systems are not explicitly optimised for pedagogy" the authors state that they "focus on conversational tutoring because we believe that it is one of the most impactful and general use cases." This forms the basis for their LearnLM-Tutor, introduced in this paper. There's a ton of information in this paper, but a lot that is not said - the model was "trained on an offline dataset," whatever that means (they say they tried several), and no stats are reported.
Today: 1 Total: 131 [Direct link] [Share]
