Natural language processing

The next step is essentially what you're saying.

Because we have the structure already and we mix that with your preference, whatever that preference is, whether that's on the visual side of things or how it's being said, And one of the things I've done in my life is I've done a study of NLP, neuro-linguistic programming, and one of the things it teaches you is that everybody works with a particular way of processing information.

So we have five senses and you have a primary sense, so meaning if I'm hyper-visual, I need to be spoken to in a visual way.

If I'm very much in the auditory kind, I need to be spoken to in a particular way.

He's now at Anthropic.

I love his stuff.

Every time he writes, I've got to read his long, long articles, but he talks a lot about NLP and the way in which input and output in LLMs are going to be different because the output's for the human more so than the agent, but the input's going to be for the agent, not the human.

So we need to think about that.

My work has often been around early aspects of virtualization, orchestration, and choreography in kind of the early days of VMware and its competitors.

And my... work mostly for the last 15 or so years has been in distributed compute, distributed data, and a lot of it has been AI for not quite 10 years, having started working with some of the NLP work that was being done just prior to the

Google transformer papers.

Very cool.

So it's not just a voice model.

It's not just an NLP model.

It's actually an SLM being developed towards context.

So even within Snorble, I'll show you a couple things.

We think of it as like a little buddy for kids because of the persona and how connected they get to it.

Now,

on the software side, we use NLP, but our AI stack is built around general AI.

At some point, we're going to tap it in when it's safe.

to Will's point there, I think two places that we've already explored using this is, first off, there's a ridiculous amount of cyber threat reporting out there, more than, honestly, anyone outside of large intelligence community assets can possibly intake.

One of the places I think that generative AI helps there in document summarization is asking, what makes this unique? Is there anything in this report that's unique relative to the other reports that we've read with NLP? So it's allowing us to ingest more or analyze, I say, analyze might be the wrong word there, but to consume more intelligence to hopefully find additional wheat while separating that from the chaff.

I think the second big place is doing some of that feature extraction, right? If you want to call that document summarization, great.

but pulling out the most relevant indicators that we can around TTPs so that, again, we're identifying those quickly and removing any overlap between the multiple reports.

So it's like giving a loaded weapon to a toddler who only wants to hear you cheer.

The NLP processes the transcript.

Kill us all.

The audio sensors register laughter.

Actually, I studied s-- um, started studying sports science and then switched to computer science in my bachelor's.

Then I did specialization in data science in the master and then PhD in NLP.

The PhD was focused on legal NLP, like ma-mainly evaluation.

The project was about anonymization of, uh, people inside Swiss court decisions.

And then after the PhD, I worked for Harvey for a year as a research scientist.

So like really hard, really hard, like long-form reasoning questions in the law domain.

Um, another one, uh, I've touched upon briefly before was, uh, called SwilTraBench, where we, um, benchmarked translation models on, on law translation and, uh, head node translation from Swiss court decisions.

And then, um, made some before on like general, like, like back in the days in, uh, in NLP when we looked at, um, named entity recognition tasks or text classification tasks, um, for multilingual legal datasets.

And then the

chatbot didn't take off, but the open source NLP library they built alongside it did that became Transformers and the company pivoted to the model and dataset hub it is today.

Huh.

I didn't know that.