ML != GenAI 🤪

Somebody shared this image on Bluesky (I'd link to their post but it's not available to public access).

And, aside from it being an interesting social experiment, I have absolutely no idea why it's relevant to anyone. You could also do an article "48 Hours Without HTTPS" about how it's hard to find websites with HTTP access, or you could do "48 Hours Without Non-GPL Software" which would make Richard Stallman happy. At any rate…who cares?

At least the subtitle does express a distinction between "machine learning" in general and "generative AI" in particular, but that's not just a casual distinction. It is everything.

I have no problem with ML. I think ML is great. I am very happy that when I open up my Photos app and search "mountain" I see pictures I took of mountains. I'm very happy that when I type "out of" in a text box it suggests "the blue". If Netflix wants to recommend something to me based on analysis of my previous watch history on Netflix, OK whatever. Things like that are perfectly fine.

What is not fine is treating LLMs like workers, like artists, like creative minds who can produce novel output. At best, an LLM is only good at providing extremely fuzzy search results through a vast content space. There are valid uses cases for that, only they're very niche and probably only helpful for expert data scientists and other highly-skilled technical people. There also need to be strict rules around the direct sharing of LLM output data as it's almost certainly remixes of illegally-obtained copyrighted material.

That's at best. At worst, LLMs and related genAI technologies are, well…what we have in the tech industry today. It is anathema to be abhorred, rejected, and fought against with every fiber of our being.

It's really crucial we keep the separation between "machine learning" as a valid programmatic discipline and "generative AI" as a crap marketing term very, very, very forefront in people's minds. Because what happens is when you say, hey, y'know, genAI sucks, somebody's going to be like "well it's going to detect your cancer so you're an idiot!" And the problem is they're the idiot because they're conflating vertical industry ML deployments, which are nothing new at all, with Sora. We need to be really, really, really careful and deliberate with our messaging and keep hammering home how one is not like the other.