Introduction

So today I'm going to, first of all, hi, my name's Natan. I'm the founder and CEO of an AI company called Anote. I'm just really grateful to be here today.

Today I'm going to be giving a talk about some of the stuff we've been building at our company, charting the emerging AI frontier. The theme of this talk is actually this pale blue dot theme.

So the idea behind this is that when I went to school at Cornell, there was this really awesome astronomer. His name was Carl Sagan.

And he essentially had this analogy of the earth being this pale blue dot saying like, hey, there's so much more that's out there than what we know.

And one of the reasons I'm really interested in AI is just I'm just curious about how the world works. And I feel like we kind of think about all the stuff we don't know and all that we can learn. It's really humbling and really inspiring.

So that I'll go ahead.

Charting the AI Frontier

So a little bit about our company.

The Mission of Anode AI

We're Anode AI, and our mission is to help make AI be more accessible, So to get some context, AI has been advancing really rapidly over the past five, 10, 15 years.

You kind of see everyday models like GPT, BERT, Claude, Gemini, Sora, all this stuff that Tal was showing. And despite all this advancements and instead incredible innovations, a lot of people are falling behind.

So even in the USA, only 15% of people use ChatGPT that are like adults. So as this technology kind of continues to progress, we need to make sure we fill the gap between the amazing potential that AI has in the everyday tasks people care about. And that's kind of why we help with these events.

Innovative Products and Applications

So our company has a few different products, and I just wanted to kind of go over them, not to kind of sell anything, but more to kind of show the potential that AI has.

Autonomous Intelligence

So the first is really called autonomous intelligence. And the idea behind this and what we're trying to build are these smart multi-agent AI systems that essentially can use a series of tools and call a series of tasks to do these awesome things together.

So for instance, if I was asking a question, can you tell me how autonomous intelligence works? The steps of kind of doing that might be navigating a website, looking at a GitHub, extracting information, searching the web, you know, formulating an answer. And those are just a series of steps.

And each step you can kind of think of as like an agent that has a task and they'll come together to do these tools or workflows, which is really useful for the end business case. There's many examples of how this has kind of been applied. I just wanted to show you an example of what this could actually look like in theory.

So in the audience today, we actually have a really awesome person. So Tony, I'm not sure if you guys know Tony, but Tony is essentially one of the most popular people at Cornell. He's a major Cornell alum, and he has a really awesome LinkedIn following, like 710 mutual connections.

And I just wanted to kind of give him a shout out for coming to NY Tech Week. So this is actually some of the people here in the audience. We have Josh, Ben, Carter.

And I want to show you how you can use like a multi-agent system to actually generate an image of Tony, right? So I might ask this multi-agent system, can you make a similar image for my friend Tony? a Cornell enthusiast. Make it like Picasso. Let's see what happens.

So what's happening here is you have this multi-agent system that's really trying to understand things. It's reasoning. It's coming up with a series of steps to search the web, go from LinkedIn.

You see some code generation on this front end. It's kind of having some logic. And then there's a series of agents actually going, scraping the internet, getting information, using this all to actually make these really cool images.

And you can see, while we're waiting, you can see some of these things that it's kind of made previously. So here are some of the people. I'll just actually pull up my friend Carter, who's here, just to give him a special shout out.

See MindStone, CIBC, him. It's pretty cool, right?

And you can see what's happening now is it's actually generating the image of Tony. And you can see all the steps that are happening. Sometimes it takes a little bit of time.

And also, if you want me to generate one of you, just come find me after this talk, and we can generate a cool New York Tech Week talk. Yeah, you can kind of see Tony. It's pretty cool, right?

So you got Cornell, Big Red, New York Tech Week, Anode. Looks a little different, you know. But yeah, that's multi-agent systems. And one of the things like we're personally really interested in our company, building something like that.

Multimodal Open Anode

The second really edge case idea that we're really interested in pursuing now is what I'd call multimodal open Anode.

So essentially, you can kind of think of Anode as an MLOps platform that helps you label data, train and fine tune models, make predictions, and evaluate the results.

And we've done this really successfully for a lot of customers like, well, Cornell, Harvard, SMP Global, NIST, Marubeni for text data. And a lot of people have been asking us, hey, can you do this for different modalities?

So things like images, where maybe you'd want to actually detect these really awesome boxes of different fishes undersea, just a random use case. Where our initial model might not work that well, but you might want to do fine tuning and evaluate and do these different training runs.

Or even things like audio. So you can kind of see, I made an Arsenal song.

yeah so like stuff like that like how do you annotate data for that how do you do this for videos like vo3 and then ultimately where people are really interested in is how do you like do these annotations and evaluations for agents so labeling evaluation across modalities to get more accurate results

Evaluation and Benchmarking

The third area that's really of interest now, that's super hot topic and we're always open to ideas is if you want to build these awesome multimodal AI systems, how do you have the valuation infrastructure necessary to make sure that the output of these models are good? Right, it's one thing to have a really awesome AI marketing video, but it's another thing to like blindly trust it when you're posting it on your social media. So you wanna make sure that you have benchmarks to be like, hey, this is good, I know it's good, this is why I think it's good.

So we're doing things related to like a model leaderboard, benchmarks for agents, creating synthetic data to help train and fine-tune models, and we are trying to figure out the best way to evaluate the performance of these models, and we're trying to build out essentially this leaderboard product around that.

Agent Registry

The fourth thing is essentially really exciting. It's essentially this agent registry.

So I guess what's really exciting about this is the way that software used to work is you'd essentially used to create a lot of different apps or systems, like front ends and back ends, to build products. So for example, you might want to kind of have email marketing agent that would search for people and email folks, or a financial analyst agent that would do really cool things.

And what we started to see is there's all these generalist AI agent frameworks where within a simple chatbot, you can kind of build these agents to do all these use cases. from random things. So if you look at like Manus AI, you can kind of see some really cool things, right?

Like this is an example of a course for how to learn quantum computing. And if you wanted to just kind of create this website, all you need to do is like view this replay, and you can actually like run all the code to... to basically see how end-to-end these agents work.

So we're trying to create something similar that has like browser use and can do these things in a way that's easy to use and easy to replicate. And you can kind of see it's all like working, right? And anyone can just try it out.

And so if you kind of believe that these agents can do all these awesome tasks, then the question is like, what is society gonna look like, right? Rather than kind of building all these systems, you're gonna have these agents doing these domain specific use cases. And so this is really where the fifth thing comes in and why we're really proud to just be here today.

Collaborations and Accessibility

We're trying to build this thing related to a product we're calling Armor in collaboration with MindStone and CIBC, where we're trying to help make AI more accessible. All this incredible AI technology has amazingly positive potential, but has significant risks to a lot of the current existing white collar jobs.

So what we really want to do is to help make this technology more accessible, make sure people aren't falling behind, teaching people how to use this technology and helping them stay ahead of the AI curve so they're able to more rapidly adjust and adopt to like all these kind of trends and use this stuff within their existing lives.

So yeah, I'd say we're really fond of MindStone, too.

MindStone has these amazing events, and we definitely recommend checking them out. They've built this awesome community.

And I think we're really just grateful to be here and trying to work together to solve this really pressing problem and learning in the process.

Conclusion

Thank you.