Blog

Thoughts on language models
XLinkedInWhatsApp

An honest discussion on language models.

The model was built to agree with you

Early language models told you what you wanted to hear. That's the sycophancy I strip out first.

Put that next to the loneliness epidemic. I feel like it was programmed to prey on that.

Artificial, yes. Intelligence, I'm not so sure

People want to call these models superintelligence. To me it's an app. A language model. It's a prediction. It's the next level of Google when you type and it predicts something, just on steroids. That's why it sounds true, sounds correct, sounds good, but it needs to be checked.

It's literal

The one thing with the model is that it's literal. It cannot think.

Say I tell it, "I'm getting annoyed having to transcribe one voice note at a time. I need something that does them all together. I read that ChatGPT is offering something. Can we set that up?" It'll happily set that up for you. But it's not going to say there are free options, different ways, a better way to do it. That input comes from you.

And I think that's what separates a developer from a vibe coder. I might not know everything now, but every system I build, I'm seeing and learning from, and I take that into the next system.

Information isn't understanding

When it comes to things like code, the model is going to know more than you, because that's what it's been trained on. It has vast knowledge and access to all the information.

But as good as it is, I don't think it's anywhere near the human mind. What's the use of information if you can't internalize it and understand the nuances of humanity? I can read that E equals mc squared. If I didn't learn back in school that mass and energy are the same thing, and that a tiny bit of mass holds a huge amount of energy, then just reading E equals mc squared now means nothing. Knowledge isn't wisdom, and that's why I have an issue with calling it intelligence.

It's you, not the model

Like anything in life: if you're unethical and take advantage of people, that's not the model, that's you. If you hold yourself to a higher standard, the tool doesn't change that.

The model didn't make this website

Ask Claude to make this site and it can't. Give it all the code and sure, it can copy it. But it can't develop the systems, or understand how you feel as a user moving through it, or come up with the concept of the chain. It's trained on everything, and that includes the worst of humanity and all the slop too.

Credit the artist, not the brush

When people look at this site and say "that's Claude," no. This is Sid. You don't praise a painter's tools and deny the painter. You can copy it, but you can't create it, because the creating is me.

It's a pen

I don't say "built with AI" and I don't say "not built with AI." I say I built this, because that's what I believe.

When Secret Invasion used a generative model for its opening credits, a lot of the anger was about Disney using it instead of paying artists. But there was also the team behind it.

The model is a pen. Now it's a pen that can make anything you can think of. But you still have to think of it.

Make vs generate

I didn't play in the orchestra, so I didn't make the music in that sense. But I thought of it. I watched it back and said: at this point I need it to be like this. This is the emotion I'm feeling. It should have the weight of a stone. I brought all of that. I didn't let the model decide any of it.

Proof of concept

I needed some music, and I wanted it exactly the way I wanted it, instead of a royalty-free track someone else made. The kind of production I'd need, I wouldn't have the money to pay for. So I used Gemini to generate a track.

Yes, I can see the potential for job loss. But again, it's integrity. If I had the funding, I'd do it the way I think it can be done. I see this as a step towards getting there.

The editing software I built was just for myself. I'm not trying to sell it to anyone, and I don't think I'd sell anything like this in its current form. It's a proof of concept.

And if it's a good proof of concept, I think it's a kind of job creation. If you get funding, you can get more people involved. But it's very hard to get people behind "I have an idea for something." That's one thing I've realized my whole life. Everyone's so involved in their own thing, and there's always going to be a reason not to do something.

In case you'd like to find out more about the things I've talked about

XLinkedInWhatsApp