Recent Experience with AI

Thanks to the generosity of Google, I am using Antigravity for free as a student. In the last week, I was using it to create beamer slides with a structure by me, to come up drafting strategies for DOTA league games, and verify the functionality of open source projects.

The biggest difference, compared with previous probability models, is GenAI’s explanability. Instead a simple multiple choice or “confidence” number, the output of an GenAI is nature language report to be examined and understood by humans. Therefore, the human’s judgement decides the action. Analogously, GenAI provides a proof script of a theorem in ITP instead of a boolean result, yielding the confidence to the verifier.

In DOTA, traditional machine learning assistant like DOTA Plus just gives a “confidence” about a hero pick, without any reason, and we have to use it as some authority. On the other hand, GenAI can explain their strategy and is open to questions. In my last league game, I started with letting GenAI draft a strategy using the link of the teams. Then, after reading the report, I critised it with my own knowledge and got the strategy modified. By discussing with GenAI, I feel more clarity than mist.

In a broader view, I often hear people complain how AI makes people dumb because of the lack of thinking. That may usually be true, if you let it do all the homework as a ghost writer. However, if you take it as your advisor, consultant and adversary, you grow more.