Now I do wholeheartedly believe, in a piss drinking competition, no other human could beat Elon Musk. So I was over the moon when Grok confirmed my thoughts, AI’s accuracy is really scary sometimes.
But then Grok started to make other claims, such as: picking Elon over LeBron James in terms of fitness, choosing to vaporise the nation of Slovakia to save Elon’s brain and claiming Elon would beat Jesus Christ in a fight. Now come on, Bratislava has an amazing chain of party hostels, it would be a real shame for Aussies in their early 20s if we lost them.
At this point, maybe just maybe, I thought AI might be getting manipulated.
It’s been exactly 3 years since the release of Chat GPT, and since then it’s been endless talk of AI being able to do anything and everything. But we aren’t any closer to blindly trusting it.
Remember when Air Canada’s customer service AI chat bot offered a discounted flight randomly? What about the lawyers in New York making up cases because Chat GPT told them? (And one for the software engineers) How about when an AI agent deleted an entire production database after being told not to?
Yes, these are all big name examples of AI getting thing wrong. But as someone who uses AI non-stop at work and at home, it gets things wrong all the time.
Theo from T3 recently did a video ‘I’m getting really tired of AI art.’ where he makes the point how AI art, music and videos are all pretty crap. They produce a final product, not its constituent parts. In coding AI speeds up development not because it writes all the final code, but because it writes some of the code very quickly. It takes you from 0% to 50% complete very quickly, but then the human comes in to perfect it. Theo highlights how the real use case of AI is when its a tool. Instead of generating an image, how about a photoshop file with layers of the constituent parts? Instead of a finished song, how about an Ableton file with the instruments separated over different tracks?
AI can’t be trusted.
AI is merely a tool to get you started quickly on an activity, to point you in the right direction. And don’t get me wrong. I love AI, I’m able to achieve so much more in the same amount of time when using AI. But hey, thats still me achieving it. That’s still me taking the completed jigsaw AI has given me, throwing some pieces out, moving some pieces around and finding a couple down the side of the sofa. And thats how a project is completed.
For software engineering it’s becoming clear that AI is just a developer tool. Like IDEs, Git and your annoying fully customised Arch Linux setup. They’re all tools to speed you up.
But back to the point, AI is just a tool because AI can’t be trusted to create the desired output accurately or correctly. For software engineering AI also poses a massive security threat!
With Grok’s recent escapade, it’s clear the ease for one of these AI companies to change their initial context. That’s a massive security vulnerability when you’re calling their APIs, the context could change under your feet and suddenly it might lead to your AI agent running malware.
No AI is not useless, but also no AI won’t eliminate 50% of jobs. Its just a very useful tool which will speed up a large range of activities, from learning about Con Thien to trying to understand why you’ve stayed up till 2am writing this.