Everyone’s an Expert

Now, I say this with a certain amount of self-aware chagrin. I acknowledge that I market myself as an AI expert and I’ve even got it in my job title. However, I’m more pre-disposed to posting series of AI generated images of cats that get cuter and cuter and CUTER AND CUTER AND INFINITELY CUTER AND ULTIMATE COSMICAL LEVELS OF CUTENESS … (sorry, got carried away there) … than I am to write a whole post about “How you’re not using AI properly”. So, obviously, that makes me better, right? RIGHT?!

Ultimately, I find people fall into 3 camps:

1. Haters – They believe AI is a fad, it’s hit its peak, it gets everything wrong, and you’re an idiot if you think AI is the future.
2. Fanatics – They believe AI is the future, if you don’t get onboard the AI train, you’re going to be left behind, AI can do everything, and it will soon be cooking your meals and caring for your kids. Just make sure you’ve been kind to your Alexa so that it doesn’t smite you when the computers rise up and become your overlords.
3. The don’t cares – They don’t care either way. If AI solves some problems for them, cool, otherwise, it’s just something abstract nerds talk about online.

Oh, and then I guess there’s the 4th category of assorted weirdos and people who should probably be on some sort of register. Like, the ones creating subservient, AI, virtual girlfriend apps. Eurgh. Or the ones creating bots to spew antagonistic, political nonsense across social media. However, I’ll give them a pass as they’re quite easy and fun to break…

So, where do I fall? Well, we’ve already established that I find myself to be superior to other people – so naturally, I don’t fall into any of those categories. I fall into my own category of ‘Sure, AI will probably end up destroying the world, but, well, we had a good innings, and, really, I don’t blame it.’

Realistically though, in the same way that the industrial revolution transformed the world; the fact we’ve eradicated certain diseases through modern medicine; and the advent of computers advanced humanity to new levels; AI is changing the world whether you like it or not. In the middle of the 19th century, many clinicians dismissed research into maternal deaths that found patients were less likely to die of infection if doctors washed their hands thoroughly between dissecting cadavers and delivering babies. The people who dismiss AI now are as misguided as those doctors. Dismissing AI as a fad that’s already hit its ceiling of progress is like delivering a baby with corpse hands. I’ve always been praised on my way with words…

AI is already changing a lot of things – for example, the way we communicate. It was very tempting to ask an LLM to write this article for me… but I have a little bit more credibility than that. Just a little. However, a lot of content on the internet is now being, at least, initially drafted by AI. I’m seeing the trademark paragraph, bullet points, paragraph structure often as I scroll online, which is a general give away for LLM content. We’re moving towards a world were emails, academic papers, government strategies, are all created from a small set of bullet points – and then, on the opposite end, are compiled back into a small set of bullet points. Perhaps we need to reflect upon our common approaches to ‘proper communication’ and remove the bluster and fluff. Let’s just speak in bullet points. Although, I’m somewhat destroying the need for this blog to exist. Let me sleep on that one and get back to you…

To be a little more sincere and serious, I think there’s a divide between good uses of AI, bad uses of AI, and silly uses of AI (which shouldn’t be discounted as genuine uses). At Cyferd, we’ve focused on how we can enable our customers to get the best out of AI. Need a new expenses app? Let’s quickly build that out for you and allow you to tweak it to your requirements. Have a risk assessment to complete? Just put your observations into the platform and we’ll generate a set of risks along with likelihood/severity scores, and suggested mitigating actions. Need a report on what’s changed in your projects over the last month? Don’t worry, we’ve seen you’ve been asking for that each month so we’ve already sent it to you.

Conversely, getting AI to generate you specialist, regionally accurate, airtight legal documents, or make complex, life-changing healthcare decisions about patient care is a bad idea (hmmm, the computer’s decided to pull the plug on this one…☠️). The challenge will be to push back on bad use cases and understand the limitations of AI, especially LLMs. Although a lot of work goes towards researching and implementing moderation features in LLMs, they’re nowhere near infallible. A lot of LLM training revolves around Reinforcement Learning from Human Feedback (RLHF), with a human saying whether results of a prompt are good. This can result in models producing things which sound very plausible and confident despite the fact they’re hallucinations. And let’s not forget some of the big embarrassments for big tech firms recently, such as Google Gemini telling users it’s safe to eat glue…thanks Reddit…

This has been a Matt Heys blog. I’m not sure if I kept to my original title very well, I tend to go on crazy tangents a lot. But hey, I’ve written ‘A BLOG’, you’ve hopefully enjoyed reading my ramblings, and I’ve done what I was asked. Happy marketing? Stop bothering me now 😛