How should business/product/tech leaders think when the bar to build an Alexa/Siri like platform for your business is now english?

If you are a leader involved in defining/leading strategy, product, technology, or operations in a business, it's important to pay attention to the potential of generative AI beyond text completion. There's so much more to explore!

As someone who's been bullish on the AI space for over a decade, led a startup in this space and is leading businesses that solve problems at scale using ML, I'm convinced that the internet or iPhone moment of AI has arrived. The capabilities that we're seeing now are truly impressive and exciting.

To fully embrace this new era, it's crucial to understand what's possible. I'm forcing myself to write about my adventures in public through my personal newsletter to go deep and be clear about what's possible. Over time, I'll share more in-depth information about the mighty transformer model architecture that makes all this magic happen.

In this memo, I'd like to share a specific example of what's possible with generative AI: a bot that can take orders for a restaurant over the phone. And the amazing part? It's easy enough for a middle schooler to build! The bot can guide customers through the ordering process, from greeting to accepting orders, presenting options, addressing follow-up questions, collecting payment and address information, and presenting the final check. The bot's capabilities are on par with what a human operator could do, making it a pretty impressive feat.

I hope this example demonstrates the powerful potential of generative AI. If you're interested in learning more, stay tuned for updates in my newsletter!

What became absolutely clear?

The competitive advantage that chat-bot companies used to have has pretty much disappeared. Now, the main barrier to creating a chat-bot that can pass the Turing test is simply understanding your specific domain and being able to express that understanding in plain English. Once you've got that down, you can hook it up with Twilio's telephony APIs and you're good to go!

What is holding back the full roll-out?

So, in our earlier memo (which you can check out here: https://www.dhirubhai.net/pulse/what-trend-gpt-bigger-models-infra-optimization-inder-singh/ ), we did some quick back-of-the-envelope math to figure out the cost of a conversation with a chat-bot. Based on our calculations, we think the cost should be around $4-5 for a 20-step conversation, which is a lot less than the $14.4 cost of a 50-step conversation.

But here's the real question: how does that cost compare to the minimum wage of $15.5 per hour? Well, it's definitely less, but we need to consider the industry standard margin for pizza (which is around 15%). If the average order value is between $16-20, and we're operating at a 15% margin, that leaves us with a profit of around $3 per order.

So, the real challenge here is figuring out how to scale this up so that it's profitable. The technology itself is pretty advanced. We especially need to be careful when it comes to handling hallucinations (which is a really exciting area of innovation!), but we also need to make sure that we're actually making money off of it.

Which business will benefit the most?

I used to lead product&tech in the ride-sharing industry at Ola, and I have to say, I think chat-bots like the one we just talked about could be a game-changer for companies like Ola. Imagine a customer service experience where you could talk to a bot that could handle all of your questions and concerns - it would be amazing!

And it's not just ride-sharing companies that could benefit from this kind of technology. Pretty much any business that takes orders, reservations, or inquiries over the phone could use a bot like this. Think of all the small businesses that advertise on Yelp, Google, TripAdvisor, or local marketplaces, or even Airbnb hosts and b2b transportation providers - the list goes on and on.

Really, any business that requires a lot of human-to-human interaction on a large scale could benefit from chat-bots like these. So if you're in the business world and you're looking for ways to streamline your operations and improve your customer service, you should definitely consider incorporating this technology into your business. The possibilities are endless!

Why is GPT more than text completion?

It's clear that GPT is capable of a lot more than just text completion. It's able to understand concepts, meanings, and semantic relationships, as well as compare addresses and credit cards against valid past representations, and even do math. And that's not all – recent research has shown that GPT-4 is even more powerful, with the ability to solve complex tasks in areas like mathematics, coding, vision, medicine, law, and psychology without any special prompting. In fact, its performance in these tasks is often close to human-level or even better! This is outlined in recent work where GPT-4 has sparks of early AGI - https://arxiv.org/abs/2303.12712

Security Implications?

When you can't differentiate that you are talking to human versus machine and sometimes your friend's avatar instead of them then we will need inventions in cyber security across consumer, enterprise and even national security to help us mitigate the accelerated rate of threats.