Jamie Bartlett

It's like the modern Latin that the church goers weren't ever allowed to learn.

These language models are actually quite good at translating that into language that everybody can understand.

People who are neurodiverse find these models very, very useful often because they're able to take existing text and put it in language and format and style and tone immediately that is accessible to them.

So there's a way you can imagine a world of these models which makes language more accessible to people.

One of the very first viral prompts

when ChatGPT was released was, can you please explain quantum physics to me as if I was seven years old?

People want to understand these ideas, but they can't.

The language is too complex.

So it allows people to learn things in ways that might make sense to them.

That might not be enough for the costs associated, but there are good uses.

And I can foresee a world where, look, if you're using ChatGPT as your therapist, that is very, very dangerous, and I talk about that.

But people are developing what you call small language models

sort of built on the big ones, which are a lot safer, which are trained specifically on sort of gold standard therapeutic data.

Clinics with professional psychologists and their patients, thousands of hours of that go into these models.

And some research is showing that these small language specialized therapy bots could be as good as you seeing a human therapist.

There are like hundreds of millions of people that need various types of mental health support, can't get it, can't afford it, it's not available.

It is not impossible that everyone would have access to gold standard therapeutic support for practically no cost or very little cost.

That would be amazing.

That's not going to happen if we all just use chat GPT all the time for everything.

But it's a way that language models could actually help certain people.