Why Episode 10?
Yes, I know, after #0 -> #1 should normally come next. But I'm a George Lucas fan. Does that count as an explanation? If not, send me a DM and I'll tell you more; no need to spread that all over an AI newsletter. Now let's get straight down to the nitty-gritty.
Why all this AI hype, and is it justified?
Bear with me, I need to go back a bit: one of the most brilliant thinkers of our time, Alan Turing, defined (among other things) the Turing Test 75 years ago. Those of you who've been around a while will remember earlier attempts at AI, some of you probably in the form of “chatbots” that popped up here and there on the websites of larger companies to say Hi. None of them could pass the Turing Test. Until 2020, when OpenAI released GPT-3, which did pass the test, but still talked so much nonsense that nobody, me included, took it seriously. The folks at OpenAI didn't give up, though, and in December 2022 they released version 3.5 in a simplified chat version, and lo and behold, it could give very plausible answers to a lot of questions. Suddenly this technology made the leap from the techie underground to the mainstream, because it had become far too easy to deal with a highly complex technology the way you'd deal with a young student. OpenAI had also taken to heart the crucial “Keep it simple” principle that had already made Google great. They stripped all the complexity of such an application out of the interface: just one input field, nothing more. No explanation needed, everybody gets it. And just like that: 100 million people recognised the potential of this technology in no time at all.
Which brings us to that potential. And it is truly huge. In my career as a digital transformation consultant, I have rarely seen such a powerful technology. It's a quantum leap, even before the quantum computers arrive. Not just because the technology is powerful, but because, as I said, it's so easy to use. Most of what made earlier transformation projects difficult and drawn-out simply disappears. The change management needed to make users understand why and how they should use a tool, and how it can improve their processes, is massively simplified. You don't need to explain AI much; it can explain itself. And it can explain, phrase, calculate or draw just about anything. By now it can also handle certain tasks on its own, and in many areas faster and better than a human. This hype is definitely justified, and it has to be taken seriously. But it also means change management has to be approached in a completely different way.
Where are the difficulties?
First: it's catching us unprepared. AI is a so-called “game changer”, except that so far very few people, if any, know how the “game” is supposed to be played. There are no rules yet either. Even the people who built it are warning us about it. It turns so much upside down that it's hard to even see where to start.
Second: it's still an immature technology. It still eats far too much energy and, above all, it isn't really reliable. That's not something we usually expect from a machine; more from people. We (probably) know how to deal with unreliable people, but this is something new. How do you deal with a technology that not even the manufacturer can fix when it delivers broken results? And who's liable for that?
Third: it's a nightmare in terms of data protection and cybersecurity! If you're not careful, the LLMs soak up everything you feed them. Personal data, trade secrets, copyrighted content, all of it can be assimilated directly by the providers and passed on somewhere else later (Funny: even Microsoft fell for it 😀 ). This is certainly one of the biggest personal and business risks to weigh up before you go prompting around too casually.
And fourth: I have rarely seen such fierce competition. Worldwide, no less. With such fast product cycles. One day a US provider brags about Thingamajig 1.3, the next day a Chinese company shows up with DeepThingamajig 3.0. And prices are tumbling daily. Imagine having to put this out to tender for procurement: they buy it, and the following week prices drop or the chosen provider falls behind technically. How do you even write a tender for something like that?
So what now?
Don't worry, we'll tackle these questions in the next issues (aka -> subscribe to the newsletter if you haven't already 😉)
So stay tuned 🙂
🚀 Quick win of the week
Define a precise context that's close to your area of expertise. In other words: put the AI tool in your shoes before giving it an instruction that matches one of your usual tasks. E.g.: “You are an experienced ... specialising in ...., make me .... ”
Enter this prompt several times in a row, do the same in different AIs, and grade the answers. That way you can judge which AI works most “correctly” in your field (Careful: because the competition is so fierce, you need to repeat this test regularly)
⭐️ Favourite of the week
https://www.napkin.ai/ I love playing around with it. It's still in beta and free (probably not for much longer), but I'd happily pay for it.
🗞️ News of the week
Meta (formerly known as Facebook) has cheated (yet again): their AI isn't as good as announced with so much fanfare after all. When you consider how much personal and professional information people hand over to this company (Facebook, WhatsApp, Instagram & Threads), this ethical deficiency is pretty worrying. When VW did something similar, there was a global outcry. Apparently that doesn't necessarily apply to US companies.
👀 Worth reading
How do AIs respond to Rorschach tests? (deepl.com to translate)
Comments welcome 🙂
