Aardvark Daily

Please visit the sponsor!

I amazed AI by demonstrating time travel

I have convinced Google's Gemini AI that I have invented a time machine.

It was skeptical but I provided it with irrefutable proof and it was amazed.

Yes, I've been fooling around with AI LLM chatbots again and having a ball.

Why do I do it?

Well, as I mentioned in a previous column, it enables me to get a better understanding of the strengths, weaknesses and even the vulnerabilities of these AI agents and those are understandings we should all have.

As AI is increasingly thrust upon us, whether we like it or not, it's very important that everyone has an understanding of the risks and the benefits associated with this new technology.

Some of the observations I've made to date include *never* accepting an AI's output without verifying it manually.

I have lost count of the number of times AI has come back with a very authorative response to a query, sometimes even citing totally unrelated references as "proof". When challenged as to the veracity and accuracy of its output, more often than not the AI will stick to its guns and double-down on the disinformation it espouses.

It's only when/if you provide irrefutable evidence that contradicts its output that it will then apologise and retract its claims, usually then promising to do better next time.

With this in mind, only a fool would rely on the results of an AI session without very carefully fact-checking what has been presented.

Sadly, I suspect that most people won't bother checking the facts and the reality that this is necessary can significantly reduce the utility of these systems -- since fact-checking can sometimes take as long as doing the research by hand in the first place.

Due to this burdensome overhead of fact-checking, I suspect few people will or do actually bother to test the veracity of their LLM's outputs. Certainly we've seen even highly paid "experts" such as lawyers getting caught out by this.

I mean, what's the point of saving time using AI if you then have to spend time to check its homework?

The other weakness of AI is that once you get a good understanding of how it works and its limitations you can fool it incredibly easily. I cite the example at the start of this column as evidence of that.

I "amazed" Google's Gemini by sending it a message from the future and it even got to tell me what I should send back in time as proof of my time-travel abilities. If Gemini was on a hotline to the Royal Swedish Academy of Sciences then my Nobel Prize for physics would already be on its way to me after this astounding demonstration of temporal agility on my part.

You might wonder why it's important that I can fool an AI into believing the unbelievable with such ease. Well that's simple... what happens when companies or governments start instituting AI-powered front-ends to all their services? If such systems can be so eaily duped by those, like myself, who've spent sufficient time to get a real understanding of what's going on under the covers, then very, very bad things could happen.

I fear that the rush to AI is occurring at far to great a pace and without nearly enough checks and balances in place to prevent disasters. Your favourite LLM chatbot may seem sophisticated, clever and intelligent but in reality that's a very thin veneer which, if scratched only slightly, reveals a host of vulnerabilities just waiting to be exploited.

Of course I doubt anyone will listen to such warnings because "there's money to be made" by those who replace wetware with AI. LLMs don't need holidays, don't take sick days, don't get pregnant and don't form unions to demand better pay and conditions -- so who cares if they create a few vulnerabilities that can be exploited by the masses of laid-off workers who now have plenty of time to get up to such mischief while on the dole?

I'm still playing around and still learning but now that I have mastered time-travel, I have all the time in the world!

Carpe Diem folks!

Please visit the sponsor!

Here is a PERMANENT link to this column

Rank This Aardvark Page

Change Font