"The dangers of trusting AI..." Topic

Hi all,

After asking about Ninja, where Chat GPT did ok, I thought I would test it on a subject I know about…

My Dad served on HMS Gazelle, an American built (Auk class) minesweeper.

I asked Chat GPT what had happened to the WW2 HMS Gazelle after the war.
It replied giving information about the WW1 Auxiliary Minesweeper…

Knowing about my Dad's ship I told Chat it was incorrect, and said Gazelle's previous name was BAM 17 (B = built for Britain, AM = Minesweeper) – and Chat was able to find the correct ship:

But it STILL called it an Auxiliary Minesweeper!

I have learnt my lesson, AI is somewhat unreliable!!!

They don't seem to train the AI significantly on historical or military history sources.

"I have learnt my lesson, AI is somewhat unreliable!!!"

The question being now is "HOW, "reliable" do you want AI to become?"

Remember, AI doesn't "know" anything. It just looks for words which commonly go together. Definitely Artificial; but not Intelligent.

Where there is Artificial Intelligence, there is bound to be Artificial Stupidity

Was at a meeting about ChaT GPT and medical publishing – as Advocate notes, the program sweeps databases but it doesn't actually think – so the idea that it avoids bias is nonsense as it picks up whatever bias there is in the data it collects

It does, however, draw very pretty pictures!

Where there is Artificial Intelligence, there is bound to be Artificial Stupidity

I'm going to be using that line.

My understanding is that the current free AIs just don't spend enough bandwidth researching your question.

When I get obviously wrong answers, I tell it to double check, when they do, they get a new but probably also wrong answer.

Garbage In – Garbage Out

Wolfhag