I genuinely don't understand why some people are still bullish about LLMs

Being wrong when you are building a grocery management app is one thing, being wrong when building a bridge is another.

For those sensitive use cases, it is imperative we create regulation, like every other technology that came before it, to minimize the inherent risks.

In an unrelated example, I saw someone saying recently they don't like a new version of an LLM because it no longer has "cool" conversations with them, so take that as you will from a psychological perspective.

Use LLMs for what they are good at. Most of my prompts start with “What are my options for … ?” And they excel at that, particularly the recent reasoning models with the ability to search the web. They can help expand your horizons and analyze pros/cons from many angles.

Just today, I was thinking of making changes to my home theater audio setup and there are many ways to go about that, not to mention lots of competing products, so I asked ChatGPT for options and gave it a few requirements. I said I want 5.1 surround sound, I like the quality and simplicity of Sonos, but I want separate front left and right speakers instead of “virtual” speakers from a soundbar. I waited years thinking Sonos would add that ability, but they never did. I said I’d prefer to use the TV as the hub and do audio through eARC to minimize gaming latency and because the TV has enough inputs anyway, so I really don’t need a full blown AV receiver. Basically just a DAC/preamp that can handle HDMI eARC input and all of the channels.

It proceeded to tell me that audio-only eARC receivers that support surround sound don’t really exist as an off-the-shelf product. I thought, “What? That can’t be right, this seems like an obvious product. I can’t be the first one to have thought of this.” Turns out it was right, there are some stereo DAC/preamps that have an eARC input and I could maybe cobble together one as a DIY project, but nothing exactly like what I wanted. Interesting!

ChatGPT suggested that it’s probably because by the time a manufacturer fully implements eARC and all of the format decoding, they might as well just throw in a few video inputs for flexibility and mass-market appeal, plus one less SKU to deal with. And that kind of makes sense, though it adds excess buttons and bothers me from a complexity standpoint.

It then suggested WISA as a possible solution, which I had never heard of, and as a music producer I pay a lot of attention to speaker technology, so that was interesting to me. I’m generally pretty skeptical of wireless audio, as it’s rarely done well, and expensive when it is done well. But WISA seems like a genuine alternative to an AV receiver for someone who only wants it to do audio. I’m probably going to go with the more traditional approach, but it was fun learning about new tech in a brainstorming discussion. Google struggles with these sorts of broad research queries in my experience. I may or may not have found out about it if I had posted on Reddit, depending on whether someone knowledgeable happened to see my post. But the LLM is faster and knows quite a bit about many subjects.

I also can’t remember the last time it hallucinated when having a discussion like this. Whereas, when I ask it to write code, it still hallucinates and makes plenty of mistakes.

If it works for you...

I use LLMs. They're somewhat useful if you're on a non niche problem. They're also useful instead of search engines, but that's because search has been entshittified more than because a LLM is better.

However 90% of the marketing material about them is simply disgusting. The bigwigs sound like they're spreading a new religion, and most enthusiasts sound like they're new converts to some sect.

If you're marketing it as a tool, fine. If you're marketing it as the third and fourth coming of $DEITY, get lost.

The rate of improvement, that's why.

These are the people who give themselves self value by saying that amazing things have no value.

Maybe Freud could explain.