AI Gets Some Right, Gets Some Wrong

As impressive as AI is, it is definitely not infallible and the top 25%-30% baseline seems to be a limit. Which is not bad if you consider that you are able to get into the top 25% of a topic that you had no knowledge about. But on the other hand, cracking the top still eludes it.

This week, Gemini recommended that I transfer Gross out and put Kinsky as my keeper instead of Verbruggen. Those two decisions alone cost me about 20 points.

Fantasy Challenge

On the other hand, in the Fantasy Challenge, which is a series of weekly challenges that basically reset every week with a different set of constraints, the model has outperformed and is in the top 10% so far. So this shows that with a bigger mandate, the AI model fails. I think this is something that needs to improve as the field develops, unless we want to have various highly specific models for different use cases instead of a general performance model.

Anyway, we have a two week international break before the Premier League resumes. Let's see if we will be able to get back on track and climb to at least the top 10% which I think would be a decent result for the model.