It's just a language model. It is easily confused by things as simple as three-digit arithmetic. If your question mistakenly suggests that something false is true, it will often hallucinate impressive-sounding reasons why this is so, rather than tell you that your assumptions were wrong. It is not going to be possible to trust any answer it gives you, especially if you're not expert in the area and the answer seems correct.

> it must be replicated in the open domain as soon as possible.

I think it's GPT-3 with some training on how to follow a conversation, right? I think there are many mature open source GPT-3 competitors to choose from, many of which perform better than GPT-3, so the replication may already be close by.