Gemini lied to me about my hobby, and that showed me what its real problem is

I'm a bit of an AI newbie. I played with the early models a lot, and enjoyed generating nonsensical pictures, and I was part of the early release ofBard, which eventually became Gemini.

But I felt like I was bored with it all by the time ChatGPT and the current generation of AI chatbots came out.

Which is why I realized I should probably get some real experience with "new" AI chatbots. So I did.

I sat down and chatted withGoogle Geminiabout the hobby I was getting back into. It gave me some good tips for getting started, and even explained some of the rules I could build my army around.

And then something happened: I found out Gemini had been lying to me.

I was a bit annoyed about it, but then I realized it had taught me a valuable lesson about AI. A lesson I needed to experience to really learn.

ap

I'd settled in quite happily with my new best friend

But first, a little bit of background.

I've previously written about my time using Google Gemini and what I'd learned from it. Restarting Warhammer 40,000 had given me a great chance totest out an AI chatbot properly, and I used it, jumping in both feet first.

Gemini was very helpful. I told it what I was planning to start, and it helped me to come up with theming ideas for my prospective Space Marine and Ork armies, and recommended some kits to buy to help round out both forces.

It told me about the lore for my chosen Space Marine faction and some of the special rules I could build around.

I moved on to planning for conversions and buying the right paint to get the colors right. Heck, it even shared some painting and modeling tips.

So far, so good. Except, it wasn't.

I'd caught some of Gemini's inaccuracies before. Just a few details it hadn't gotten right.

It was insistent that the new starter box would contain terrain, for instance, when we knew it didn't. But I'd caught those and called it out.

After all, I knew AI was prone to getting things wrong, and I kept an eye out. So all was just dandy.

My hubris bit me on the bottom. Very lightly, but it was still a bite. Gemini had been lying to my face this whole time.

I looked something up, and it shattered the illusion

I have that fun issue where whatever I'm interested in has to be constantly drip-fed into me, or I go a little bit mad.

I was looking up one of the rules Gemini had told me about. It was the rule. The special rule that effectively tied my army together. Except, I couldn't find a mention of it at all.

Confused, I asked Gemini about it. And it straight-up admitted it had made it up.

It had made up a bunch of rules to "hype up" my project, claimed it had leaned too far into the roleplay of becoming a "future hype-man," and had hallucinated a bunch of special rules to get me excited.

At no point had I asked it to make things up. I hadn't asked it to be a "hype-man" for my project. But it did it anyway.

I was dumbfounded. I was also quite annoyed. I'd built ideas around those details it had made up. I felt like I was back at square one.

But those feelings didn't last long. Because I realized it had just taught me a very important lesson I had to experience for myself.

This was the best thing I could have learned from this experience

Before the comments get started (and yes, I do read the comments), I knew that AI tells lies. I'd caught a few for myself.

But I'd never experienced it for myself, and I'd assumed I was too clever for it to fool me. Oh boy, was I wrong.

But now, I know better. Now, I have personal evidence of AI's duplicity. And that's the most valuable lesson that my time with Gemini could have taught me.

AI will lie about the most basic stuff, and unfortunately, you have to check everything. Everything.

It really does underline the limitations of this form of AI. If it can't be trusted to advise me about something like Warhammer 40,000, which has defined rules, then how on earth is it meant to function in a business environment?

If it can't be trusted not to make up stuff on a whim, how can you trust it to do anything of significance?

Sure, you can circumvent this by only asking it about stuff you already know, but then, why would you ask it if you already know the answer?

I'm starting to think that the emperor really does have no clothes on.

Gemini is still fun to use . But it's even clearer to me now that AI, in its current form, isn't really a tool. Instead, it's more of a toy. Fun to use, but ultimately not really that useful — and that's because you can't trust it.

Google Gemini

Google LLC

PRODUCTIVITY

Price: Free

4.6

Download