Fraser Sherman's Blog

I ran into a bizarre, but apparently genuine post online recently. A tech guy bragged that when his wife wanted him to read their daughter a bedtime story over the phone he proposed an audiobook. Mom: no, she should hear your voice. Dad: I’ll have AI read it in my voice! Because when you work in AI, that’s the kind of outside-the-box thinking you come up with.

Next this assessment (not of that incident) by Rebecca Solnit: “A lot of what AI use amounts to is the abandonment first of mildly demanding activities and then the abandonment of your own belief you have capacity to do them, and then the actual ability to do them them. Maybe an abandonment/atrophy of natural intelligence.” She’s writing about Marissa Mayer, a techpreneur whose new LLM is supposed to save women the costs of emotional labor. In this case, a teacher asked Mayer for a list of three cool things about her kid; Mayer “likes cracking a hard problem” but for a challenge like this, she had her new proprietary AI, Dazzle, come up with answers. As Solnit says, this is undoubtedly partly a PR push for the tech — but it fits with so many other things I’ve blogged about. People repeatedly outsourcing stuff that’s within their power because it’s uncomfortable or they’re not sure they can say it perfectly and everyone knows a computer can do it better.

Wired, for example, looks at how tradpublishing is embracing AI, often without telling authors. Why have staff write responses to authors or offer critiques when you have a machine? Isn’t that better.

For another aspect, let’s talk this post:

Needless to say, I do not think the author of this post was ever writing a book And that’s said. It’s not just the ethics of pirating everyone else’s work, though that’s certainly a factor. Lots of authors start out imitating other authors. I did bad Lord Dunsany and Bad Lovecraft; I improved. Robert Bloch started out doing Bad Lovecraft and he improved a lot. If all you do is tell the LLM “turn my prompt into a Cthulhu Mythos story” you’ll never learn to do better.

Or consider Marc Andreessen’s argument that thanks to AI, someone with no set, no camera, no actors and no movie-making skills can still make a movie. That shit has never stopped a determined low-budget filmmaker: they’ve shot films in their basement, their living room, the street outside. They’ve used their neighbors or strangers on the street as actors. A lot of these films suck; some of them are great. Post-WW II Italian neorealist films were almost all shot on location with non-actors because Italy was in no shape for anything more sophisticated. And even the directors who made bad films had a chance to learn by doing. Telling your AI to make a movie — again, you learn nothing.

Like writers, tand-up comics steal from each other but there’s a limit to how much damage one comedian can do. Instagram, however, is allegedly allowing hundreds of AI comics to perform (for want of a better word) and they’re straight up stealing real people’s routines. After all, why should we get the benefits of our creativity or imagination when Silicon Valley wants it all?

In other AI notes:

Using LLMs to verify the logic of math proofs has become a game changer. Like so much else done with LLMs, the results aren’t reliable.

The hell of facing a job-application system run by computers.

“without feeling anything, without caring, they convince people that they are empathic, that people are being heard when they speak to machines. And what’s missing in that behavioral judgment of whether something is empathic is all of the things that empathy really does for a person, which involve my commitment to you, my caring about you, my being willing to sort of walk a little bit of the road with you”