For what it's worth, I've found it to consistently produce very convincing garbage. I asked it a basic question about Golang, and it bullshitted an answer that I couldn't tell was wrong until I tried it. It can't reliably perform arithmetic, and while it is capable of identifying that it gives inconsistent answers, I haven't yet got it to admit that these inconsistencies meant it was wrong. From my limited experience I would strongly recommend not using it for this.