Listen the following video: How build a company around Cutting-edge AI.
Three breakthroughs driving AI forward with Srinath Sridhar *PHD. in Computer Science Stanford Harvard, Founder of Regie.ai
B2B A CEO. A Podcast for Technical founders-foundation Capital with Ashu Gar.
3 breakthroughs driving AI forward, from Regie.ai founder Srinath Sridhar
Key ideas
/ Investors want to invest in a company when the train is just about to leave, and then everybody tries to pile on. Building the train together with the investors, is an early investment, opposite to waiting for the train to leave the station.
/ There was no specific idea or product in mind, but he had such a compelling vision for how the world would change with LMS. GPT had just lunch.
/ Computers Were not able to string two sentences together until then. The key moment was when, for the first time, computers could like literally string two sentences, and actually make some sense.
/ Text generation.
/ Natural language processing or understanding, was all hand coded, it was brute force.
/ State-of-the-art. Grammar. Understanding. It’s basically just compute the list of four word frequencies, and then a human would basically go in, and say why a series of four words is less frequent than the other series of four words, and why that’s a mistake. Thas was the state-of-the-art.
/ Today you don’t have to teach computers any grammar, you don’t have to do any of this frequency counts, and you can actually understand grammar by itself. And once we knew that it could do two sentences together, then we knew that you can scale it up.
/ What applications can we do with AI. He had a tough initial 6 months to try to figure out how we take this piece of technology and apply it to the space.
/ How do you compose an email. With Google auto complete which was an innovation. Indeed it was the latest innovation. Thas was one insight.
/ others insights: composer side. Google autocompleting, that it autocomplete only the word that you’re literally typing right now. No sentences, no paragraphs, nothing more. The inbox mail hasn’t change in 25 years. Google, Outlook… Priorizing, aggregating… were all very manual. He had the idea to improve email for sales people.
/ Your really innovate at how to compose an email using GPT 3. Looking for a problem to go solved. Creating sequences of emails should be done every day, is painful, is broken.
The initial product was sequence composition, a series of emails. There are sales engagement platforms like Outreach that would send the first email, somebody doesn’t reply, they send you the second email and so on. And all the large B2B companies today use one of these sales engagement platforms.
/ The initial product was how do we compose a series of these emails, that can actually could be exported into Outreach for them to execute (autogenerating sequences). It’s very similar to the olden days of asking a set of questions to GPT3 at that time. Instead of spitting out a blog or an ad or an essay, it would spit out a sequence of emails, these emails would just get exported to Outreach. Nobody paid any attention to it at the time. Then we built out a product to write one to one emails.
©image. http://www.outreach.io
@Yolanda Muriel 
yolandamuriel.com 