Discussion about this post

User's avatar
Kalen's avatar

I've periodically gotten myself some kind of LLM access during this whole hype cycle, and invariably go through a two stage process-

1) 'wow, it's neat a computer can do that now'

2) 'huh. that's not actually a thing I need- I'm cancelling this because everything else about this space makes me feel bad and gross'

The number of circumstances where I need slightly customized boilerplate is just not that large- and I have an 'email job'! I need to tell people things that are in my head, which the LLM cannot help with, I want to think things through by writing about them, which the LLM actual hinders, I want to sound like me, and if I'm creating something that's going to go out again and again the LLM is not good enough- any of the 'intelligence' benchmarks the model supposedly sail through these days are somehow overlooking that it only takes two or three prompts for the model to engage in some 'Clifford the Big Red Dog is not a dog because dogs are not big, except for Clifford' logical breakdown. Meanwhile, over on one side Ray Kurzweil is still wandering around muttering that we don't need to worry about anything ever again because AI will make food for us.

The one place I get any utility out of LLMs is as a search adjunct, but even then it's a close thing- Google's AI overviews actively annoy me, but Kagi has a 'summarize this page' button next to their search results that I use every once and a while to check for relevance, and, eh, occasionally handy. Notably it's slightly customized boilerplate being generated to help me think, not to communicate my thoughts. I'm building a document parsing thing as part of a work project that will probably have some kind of neural net deep in its innards, but making it do what we need is an actual, normal engineering problem- maybe it would have been flatly impossible before, but it still takes work.

Expand full comment
Leah E. Ward's avatar

“One political issue young American men are genuinely passionate about is their absolute right to lose a ton of money on an app in a really stupid way.” I laughed out loud in this Amtrak train, this observation is too real!!

Expand full comment
9 more comments...

No posts