#54 🌟 AI - An Inch at a Time.
Exploring the the Who What and Why of AI
Another great week and we’re still trying to piece together how to use AI for fun and profit. Thanks for joining us.
Who
As I write this issue, I’m thinking about myself as well as about you - the reader.
What are we doing with our lives?
Who do we want to be by the end of the year?
Who might we become if we embrace the tools and techniques the experts in the AI world are creating to advance the bleeding edge of tech every week.
A year ago, AI was AMAZING!
Words came out of the copywriting tools like a scene from Fast and Furious. And, like in the movies - most of the excitement was us thinking that we can move fast - without crashing. Alas, we often ended in an AI ball of flames and a wasted an hour realizing that AI wasn’t actually thinking at all. It was just guessing. Damn it!
But, could it be different? Slower? More methodical? Could we actually build our words or code more like Lego™ blocks?
Who will I be if I learn to build with AI?
What
This week, I’m very interested in a methodical approach to building stuff with AI.
When writing, I start with my own thoughts. I get them down then edit. I might put them into ChatGPT and ask for an enhancement. I might not. When I do, I often get insights I didn’t think of. I’ll rarely use the AI output directly. It just doesn’t feel right.
If it’s code, I try to ask for small things. AI doesn’t like to move in baby steps. It doesn’t have a good memory, so it likes to spew everything it can think of while it has a thought.
I tell Claude.ai that she needs more context and I realize, by her response, that I’m the one that needs to provide it. I need to tell her what I want her to code - and how - and maybe then, she can get it right.
I’m calling her Claude - because I like the female energy.
The updated Opus model is great, but her training data ages quickly when it comes to technology and AI. If I ask her for code from the OpenAI docs - I get outdated API code that doesn’t work. I can get help fixing it, but it’s a pain. ( and makes me realize how lazy I’m getting )
Why. Can’t. It. Just. Work? It’s AI for freak sake.
Working with AI is a lot like working with a fast amateur. They are eager and quick, but they need training. For now, we have to be more methodical and remember that they, like us, might not have the best memory. And, no matter how much we work with them, the model doesn’t actually improve until the team in the lab improves it.
The “What” - I’m realizing - is that I need to take things in much smaller steps when working with AI. Take many more steps, possibly, but also smaller steps - because of the propensity for AI to error.
Why
When I was a kid my uncle had horses. He put me on the biggest horse for my first ride, because men want to see boys grow up fast.
But, he didn’t teach me how to ride and I didn’t know how horses worked.
You may not know how ChatGPT works or LLMs or AI or other powerful, but seemingly harmless tools. But let’s agree they are powerful and they can quickly become a footgun.
After he got me in the saddle, we started walking. The horse wanted to move - he wanted to run. I must have annoyed the horse, so, he laid down and started to roll to get me off. I didn’t realize I was in control with the reigns. I didn’t feel in control. I freaked out a little and jumped off.
”Why the heck did you get off?” my uncle asked.
I said nothing.
The horse got up and started running. It took us 30 minutes to wrangle him and get him back under our control. It was a lesson I didn’t know I’d needed until AI came along.
The Lesson: Don’t let AI run away just because it can. You are in control.
You may have to work a little harder to figure out how AI and large language models respond to your commands. To get things exactly as you what you want it takes a little work, but it’s possible.
Join communities. Find prompts on Reddit, Read the docs. Find people who know more than you. And.. take it slow. Heck, even ask ChatpGPT - “How do I prompt you to get the result I want?” - see what it comes up with. Learn as you get feedback.
The AI tools are wonderful and powerful and there’s no reason you shouldn’t get back on the horse every time it tries to steam roll you with boilerplate answers or outdated facts.
Let AI do what it’s good at - and hopefully you’ll at least have a good story to tell.
And now… another great example by Zeng!
Ai gone wrong😱
Have any AI gone wrong images you can share?
😄✨ If you didn’t like the stories, we’d love to hear about them.
See you next week,