"When AI starts giving history lessons" Topic

I guess most of us have noticed Goggle searches are now often starting with an AI answer. I got this today while searching for info on the Battle of Sulys during the Hundred Years War.

The Battle of Sluys was a naval battle fought off the coast of Flanders between the English and French fleets. The English won the battle, destroying a large portion of the French fleet. The English victory was due in part to their use of longbows, which were more advanced weapons at the time. The English longbow's range and rate of fire gave them an advantage over the French cavalry, which was slowed down by the wet ground as they attempted to ride up the hill to reach the English.

I can only imagine how wet the French cavalry found it in a naval battle 🤣

Anyone else had interesting AI search results?

Knowing a little bit about some of the algorithms that were used in the early days of AI it doesn't surprise me. The starting assumption is that there must be an answer and so it will produce one. Checking mechanisms are obviously still quite crude.
In the very early days of spell-checkers I got an embarrassing result doing a demo to a class of middle aged ladies. I mistyped 'computer' as 'copmuter' and its only suggestion was 'copulate'.
Sadly I think we are in a situation where machine intelligence will be better than human because humans will get worse through lack of mental exercise.

I was looking up my Dad's ship in WW2 and even put WW2 in the question, AI (Chat GPT) insisted it was a WW1 auxilliary minesweeper!!
I put a second question giving the number of my Dad's ship (J342) and got a better answer, but again 'Auxilliary Minesweeper'
Epic fail from AI!

At this point, AI makes Wikipedia look like post-graduate research.

Well, I said it could generate second-rate fiction, and the evidence so far supports me.

And students (and I include American college students) will confidently put those "facts" down on whatever paper they are turning in to their teacher. Just like when the Germans bombed Pearl Harbor.

I saw an example where the answer cited "studies" but when pressed AI could not name/produce them.

It just new "studies show" was how you answer that kind of question!

I've been seeing horrific misstatements like this for a year or more, including entire pages in AI-generated encylopedias full of misinformation and grade-school level English.

Search engine results are now so full of LLM garbage that they are approaching useless. As long as the web is run by ad revenue, there are going to be hordes of clickbait honeypots polluting every search, and it's up to the human at the computer to verify information him/herself.

- Ix

If you understand what it *is* it's a super useful tool. If you do not understand what it is then it just looks dumb. If you *really* don't understand what it is then *you* look dumb.

But not only do people think it's something it is not, as it gets better it will appear more and more to be the thing that it isn't and dupe even more people.

I have the app on my iMac and use it every day. Mostly for coding, but also for other tech support and ideas. I don't ask it for facts I won't instantly recognize as true or false.

You *should* be able to paste in rules and ask for interpretations. My experiments with this have been very successful. If you don't paste in the rules and just ask "are you familiar with the game" then you get hallucinations. But if you paste in the rules it can analyze them pretty well. I'm not sure I *trust* it here, but it seems to do well. I tried pasting in some rules and asked for an example of play, this did *not* work.

I wonder if, just as some things make the rich richer and the poor poorer, generative ai may make the smart smarter and the dumb dumber. Maybe not. Sounds clever, though.

For fun, paste in a magic spell description from you favorite game and then say can I use it this way or that and you usually get good answers. When you ask it how you can use this spell to impress or embarrass someone or send a message or make a living you get plausible answers!

But it's not a search engine. It's a constrained random text generator.

Doing some research on the War in the Pacific, 1879 gaming scenarios – AI suggested the colorful uniforms of Chile, Bolivia and Peru with Battlefront Miniatures and tanks were a good mix.

Keep looking for this 19th century tanks that Battlefront makes… Certainly would be a different scenario…

If AI ever becomes a problem, the clear solution is to disconnect it from Wikipedia. The dang thing will go promptly mute.

Is it possible the French rode Seahorses into the battle? That would make the AI answer correct…

just saying,

Dave

Scythed chariots for Peru?

There is a simple thing you can do with your AI searches: Ask it for the sources it uses. Then you can double-check the source material if you aren't sure. Otherwise, just assume the AI is making up… ;-)

I conduct historical research and use AI as one of my research tools. I always ask it for sources and for links to those sources.