Garbage in, garbage out

And the AI tools I use are all built with what I call when I teach classes on how to use AI, the art of the prompt.

And if you go back, and now I'm dating myself again, back in the old days, we used to call it GIGO, garbage in, garbage out in computer life, you know, when they're programming.

I still say it's GIGO with AI, but now I say go for Vivo, value in, value out.

And what I do is in each one of my tools, there's no magic bullet to AI.

I ask a series of questions, except for the smiley one where you smile in your camera.

If you feed it garbage, it will output garbage.

So if you input... x you get x and even with a lower quality system you can input like you can input good data and you can get good data out right and so a common fallacy in the software engineering world is forgetting this it's forgetting that the quality of a system's outputs are really fundamentally determined by the quality of its inputs And sometimes, well actually quite a lot of the time, almost all of the time, your time as an engineer or a systems architect is better spent sanitizing inputs than it is working on fancy processes to deal with disparate sources of messy and dirty inputs, right? Because of garbage in, garbage out.

Now, why I tell you this, is because we come back to this.

So I'm doing these job interviews and I do 300 of them or something.

And I realized that This was a really poor way to measure, and it was a really poor way to interview people.

And that reminded me of this documentary called Chef's Table, where the ingredients of the food determined the quality of the meal, right? And the world's best chefs were the best sources of ingredients.

And so then that reminded me of the garbage in, garbage out thing.

And then that led me to do a little bit of research.

And so the root of the problem can be at a firm, hey, you're using AI to build some tools that do the things you want to do, and this is wonderful.

But you still have like a, you know, the prospect for garbage in, garbage out, the old, you know, the old phraseology there, like, you know, if you've got a data problem, no amount of AI is fixing that necessarily.

So how do you, when you work with firms, how do you go about helping them assess whether their data is clean and usable? Because that feels like that's the thing we need to do before we start building new stuff.

Yeah, well, one thing is that we've been doing this for a long time.

And, um, maybe start with some of the, some of the challenges that you see when they're connecting those, those wearables, whether it be a, a GPS watch, a heart rate monitor, and a, and a power meter.

Um, what are some of the, the pitfalls that are gonna mess that whole GIGO, uh, formula up for the user?

Let me think.

I think one is just, like, inputting the information correctly.

Totally.

Um, and then there's the, then there's the calibration of thresholds, right? Like, a lotta times we, we want to, um, you know, w-we wanna have demarcation points, whether it's, uh, you know, a, a peak power, uh, a maximal aerobic power, um, a threshold power, a first threshold, a resting level, right? And these can all be off as well.

And then when you're interpreting the data and those are off, well, now you have another GIGO issue.

There's so m- there's so many, right? Like, uh, you know, on the, on the one hand we l- we let off just with the miracle of A- of AI on all these things, but at the same time, there's, there's so many pitfalls.