Mastodawn

@carlton I was listening to the end of https://pod.link/1804434846/episode/6205f713029cf3d27b29c258658dab19 this morning, and I thought their points were pretty good.

Abstractions

Abstractions is a podcast about technology, software, hardware, and the Internet, and the way the ever-increasing layers of these technologies permeate every aspect of our everyday lives.

@webology JeffBot please summarize 😉

(It's not a short episode)

@carlton Start about ~10 or 12 minutes from the end.

The tl;dr is that we are getting confused by LLM's willingness to always provide an answer because of how they are tuned. Which is annoying, and that's why I prefer the results I get from the APIs directly, because I have more options available.

Strangely, I'm not sure why my results were better other than Claude hit the web to find them.

@webology I suspect your prompt was more demanding?

(That's also a point I have planned to discuss…)

@carlton I copy and pasted your question. I did capitalize the "US Masters" though, but everything were your words. https://claude.ai/share/a464d9e4-9efe-4e2b-bc21-561698c8baa6

Without the ability to do a web search, I don't find LLMs to be good for grabbing random facts.

A fun one to try is give it a list from a requirements.in file and ask it to give you the pypi and GitHub or whatever repo for each one.

@webology 👍

> I don't find LLMs to be good for grabbing random facts.

Exactly with you! But that's precisely how I see normies using them. I'm not advocating it, rather just preparing an example to demonstrate the point, to my Dad.

I repeatedly bump into bewilderment about how the output could be anything other than correct. The idea that they're fact retrieval engines seems deeply embedded with folks in the wild. 🤷 (Others have more to say on this than I do.)

@carlton Sure, agreed. I think what's changing is their ability to find/look up/use tools to find the right answer. I find that to be worth the wait. 🍿

@webology Yeah, I'm not having a pop at you in any way. 🤗

Will Vincent 2d ago

@carlton came to read whatever Carlton writes. Stayed for the golf.

Love the AI as teenagers analogy: "The trouble I have with them is that they're not experts at anything. (They're teenagers so they often think they are, but that's not the same.) It's hard to get them to see, and feel, the issues here in the same way."

@wsvincent That was about my kids 🤣 — but, yes, similar with LLMs. Overconfident bullshitters.

Andy War 2d ago

@carlton the problem with the kids is that they will never get to be experts in something