AI is a like moody teen, here's why
This post is for anyone looking for a simple way to initially wrap their head around AI, and why it can spectacularly fail, while also getting some things scarily correct. In essence, treat AI like it is a moody teenager.
I didn’t start there, given that it came seemingly out of nowhere, I initially felt like AI was more toddler than teenager. We play with them, and every time they accomplish something unexpected (even if mundane), we collectively freak out.
Of course, for every accomplishment a toddler has as they develop, they fall over tons more, even after previously standing up. We expect more from teenagers than toddlers, and our responsibilities and the way we handle their failures are different as well.
The more I play with Generative AI (GenAI) and Large Language Models (LLMs), I increasingly view their capabilities and access to a deep well of information similar to a teenager's expertise and ease with TikTok. The things it “knows” how to do, it can usually execute incredibly well.
Toddlers are terrible at communicating, but teenagers develop their own slang to say less, while still getting their point across. Learning the slang to interact with a LLM is what prompt engineering is all about. It’s optimizing your slang with the LLM to get the best possible outcomes.
Teenagers are also on the way to adulthood, but they’re not there yet, and we can sometimes forget that. There are many actions that I expected LLMs to do easily and quickly, but that it stumbled on. Oftentimes, I realized I could have accomplished the task faster myself, pulling from my own hard-earned knowledge, but instead I was going through exercises to make the LLM get it right. That is model training and tuning.
We also don’t dismiss teenagers entirely because of their failures, we give them room to mature, but with limits to prevent them from going too far off the rails. We are still responsible for instilling ethics and social responsibility within them.
Teenagers can understand these concepts, toddlers cannot. This is also why we can’t fall for the “it’s too early” narrative, whenever we bring up legitimate concerns with AI on multiple fronts.
Finally, if you’ve only played with the free versions of GenAI tools, then you have only scratched the surface. In the rush to monetize this tech, access to the most powerful capabilities are often limited behind subscriptions, early access programs or restrictive licenses, even while our data is being used to train these models.
This seems very different from the “user acquisition at all cost” approach we saw in the initial surges of the social and mobile waves, where they avoided monetization until dominant. Perhaps that's due to a belief that the dominant players are already clear, which might explain certain antics of late and that swagger is definitely not always warranted.
I hope you’ve enjoyed this AI amuse bouche. If you’re looking for a quick course to get started, I enjoyed taking this Trustworthy Generative AI course. You can even audit it for free.
I’ve hoovered up a bunch of courses and resources during the past 18 months of this new journey I’m on, covering a bunch of topics, not just on AI. In a follow-up post, I can share some of that with you.
Interested in working with me on an AI project, or my other consulting services listed here?
P.S. Happy Pride Month 🏳️🌈
