Any ideas on how we can tame the AI?

Human-in-the-loop

no idea, just going around blind until something sticks.

Well, the first thought coming to my mind was to control the training data. Any AI model, whether it's ML or ANN, gets its behavior and knowledge from data. If you want a model to behave in a certain way, just give it the data that fits your intentions, and the model will not be able to produce a result it hasn't been trained to produce. Actually, it will give you some outcome, but it will be wrong anyway. For example, the training data for ChatGPT was supervised and censored a lot so that ChatGPT wouldn't be able to give you violent, cruel, or abusive responses. You can read it here (long read): https://time.com/6247678/openai-... Also, you can control what an AI model gets as a request from a user. Models don't work on their own – they're called. So, filter the request to avoid undesired behavior.

Taming AI? It's all about regulations and ethical AI frameworks. AI's like a kid, needs boundaries to not go wild. OpenAI's doing some cool stuff in this area. But remember, it's a fine balance.