Zach Lipton

doing your paperwork for them, providing context, helping them to surface important details from past records, from the medical literature.

Like, none of that was happening.

So we kind of sat on this idea that, like, it was sort of intelligence at the point of care.

Like, this was the place to be, be inside the conversation.

Only problem is that the technology in 2018 was nothing like what we have now.

For reference, the largest language models at the time were like, you know, quaint by today's standards, like 1,000 times smaller, 2,000 times smaller.

They tended to have a look back of they could process about 100 tokens of text.

So it was like they were looking at about a paragraph generously.

A small paragraph.

So you had to figure out, how am I going to summarize a 4,000-word conversation or a 10,000-word conversation, let alone a 10,000-word conversation plus the entire medical record of supporting context?

How am I going to transform that into medical documents using the technology of 2018?

And so I kind of was like, well, sign me up for that problem because that's...

That's the machine learning problem.

That's the AI problem.

And I was like, I'll work on that.

And not only that, but like, I kind of took a posture of instead of optimizing for like, you know, you know, like sort of just like personal stake, it was sort of like support my lab.

I'm going to work on this.

I don't know if this is going to take one year or if this is going to take 20 years, but.

Probably not one year.

Support me to work on this as an open-ended research problem in the university.