Was at a tech mixer/meet-up/networking thingy in Silicon Valley area just before July 4th holiday.
“And we have a lot of machine learning in here” said the entrepreneur, indicating his app.
“It’s classifying product categories automatically based on the description and the manufacturer?” I ask.
The entrepreneur nods.
“A Naïve Bayes classifier then.” I add.
The technical cofounder, silent until now, eagerly starts to explain how he implemented it in NodeJS with an off-the-shelf gem.
“No, no, far more than that. We have a lot of machine learning in our app. This is Deep A.I.” said the business co-founder using air quotes.
I swear you could see the italics.
I looked at the technical co-founder without saying a word. He continued to babble about Naive Bayes and gems and NodeJS.
I asked about scaling the P values and if he was using a logarithmic function.
The run-on-sentences stopped. The response was slow and thoughtful. “I think that’s how I stop the values always trending to near zero. I don’t really understand that part.”
I nod politely. “Nothing wrong with that. Use whatever works, even if you aren’t sure you fully understand how it works.” I say encouragingly.
“Dude, shut up, you’re talking about stuff we’re trying to patent.” said the business co-founder impatiently.
“It’s just a Bayes classifier” opined the technical co-founder.
The business co-founder looked pained. “He doesn’t really mean that. It’s more complicated.”
I nod again, just as politely.
“Did you try a Laplacian smoothing algorithm? Or additive smoothing?” I asked knowing full well that Laplacian applies to polygonal meshes and that additive smoothing and Laplace smoothing are one and the same.
“Yes” said the business co-founder.
“I don’t know what those are.” said the technical co-founder the merest fraction of a second behind.
“Have you solved the overfitting to data problem?” I asked the business co-founder. A Naive Bayes classifier generally doesn’t suffer that problem.
“Of course. I won’t deny we had some trouble, but we overcame the overfitting issue.”
I nod again.
The technical co-founder looks like a deer caught in headlights.
“This is a very nice looking app” I said to the business co-founder. “But I will give you one piece of advice.” I didn’t look up from his phone that I was holding on to. I was still swiping through the screens of the app. “When you talk to potential investors, do not bullshit them about your technology. It’s no fun to lose out on an investment during the due diligence phase. The tech guys working for VCs are a lot sharper than I am.”
Now I had two deer caught in the headlights.