Aardvark Daily

Please visit the sponsor!

This is serious, I am not kidding

Okay, I confess, I am now addicted.

Addicted to what?

Well in recent weeks I've been experiencing the same kind of dopamine hit that I recall experiencing back when I first played the amazing "Adventure" game on an early microcomputer.

Unlike today's modern games with photorealistic graphics, 3D realism, high fidelity sound and highly complex game-play, the original Adventure game was all-text and relied on logic and determination on the part of the player.

The novelty of the whole microcomputer scene, combined with a program that seemed to understand your commands as you moved around a totally virtualised world was awe-inspiring in a way that modern flashy games just can't match.

From memory, I spent almost 24 hours with out a break before I finally completed the game of Adventure. Having done that I set my sights on Pirate Adventure as my next goal.

So what have I been doing recently that has revived that level of addictiton and enjoyment whilst sitting in front of a computer keyboard?

Well I have already mentioned the fun I've been having as I've tested the boundaries of AI chatbots in an attempt to better understand what's going on under the covers.

That experimentation is fascinating and even more challenging than those early text-based Adventure games.

I have now accumulated a significant amount of understanding and knowledge which is making it increasingly easy to sidestep the safety mechanisms that are built into these LLM chatbots.

My latest success was yesterday when I convinced Google Gemini to tell me how to actually create successful attack vectors -- and yes, they actually work.

That I was able to convince this AI to help me hack itself reinforces my belief that I'm doing pretty well at this game and if I can manage to do this, imagine the potential for actual bad actors to do the same.

Every time I am reminded just how vulnerable these systems are to exploitation I shudder because every day I see more stories like this where companies and even government agencies are rushing to embrace the power of AI in the name of cutting costs and boosting profits.

I know I've written about this before but it's really hard to believe that so many organisations are so seemingly unaware or unconcerned that they're painting huge "hack me" signs on their doors.

The fact that someone as old and worn out as myself can, after a period of probing and learning, come up with strategies that then allow me to use an AI against itself to discover and exploit vulnerabilities in its protective mechanisms to the extent that it will hack itself is insane.

If I was to relate the fundementals of just how simple such hacks are I'm pretty sure everyone reading this would be gobsmacked at the ease with which this can be done. Imagine the damage a well-resourced state-funded actor could unleash on organisations or governments that leap head-long into the roll-out of AI without some way to mitigate the risks.

I will continue what I'm doing, not because I want to hack AI systems but because the level of intellectual challenge is enormous and, as we all know, the magnitude of rewards are generally proportional to the size of the challenge.

Although the latest raytraced top-tier photorealistic computer games are of no interest to me, this new "Adventure", which requires navigating deep into the secret workings of an AI chatbot, really do re-awaken the enthusiasms I enjoyed back in the late 1970s.

Maybe at some time in the near future, I'll make a video documenting my successes and demonstrating the scale of the problem to those who are blissfully unaware at this stage.

In the meantime, I will be steering well clear of any company that might place my personal data or financial wellbeing in the hands of AI systems and I'd recommend you do the same.

Carpe Diem folks!

Please visit the sponsor!

Here is a PERMANENT link to this column

Rank This Aardvark Page

Change Font