IPO Corner: Embrace The Chatbot

AI is taking over the world right this very minute! At least, that’s what’s happening if the internet is to be believed. We don’t know about you, but rapid-fire stories about AI chatbots, such as Google’s Bard, have been pulsing through our news feeds faster than we can consume them… we admit, we’re anxious to see what they can do. And we’re all about to find out too, because we’re putting them in everything, and nobody can stop us.

Just in the last few days alone, Snapchat announced they’re making their own chatbot, called “My AI”, free for every user on the platform. It seemed to be a bit surprising to some, as MY AI is only a few months old and has only been available to a few million paid subscribers up to this point. It sure looks like everyone is jumping in feet first, like we’ve collectively decided “darn the torpedoes”…  maybe now, the AI race is truly on.

When you think of an AI Chatbot, the name that probably comes to mind at this point is ChatGPT. Which is understandable, but it’s important to know that there are already many such chatbots in existence… and more are coming online daily. So, it’s a little more helpful to learn a bit about the underlying technology. The most important thing to understand about AI chatbots is, at their core, they are powered by what’s known as a language model.

Language models are what allows the AI chatbot to “talk” to you the way a human being might. Human beings learn from their environment, they explore and figure out what things are so they can make new relationships with objects and people, and then express the new information when appropriate. That’s really all that happens with an AI chatbot. They are trained on a language model, which is basically their “environment”.

A language model is a giant set of data. To keep with our analogy, imagine a child in a daycare setting… the toys, furniture, and everything in the room are the child’s “data set”. All the child can know is what’s in the environment, and a language model is a lot like that. We give the computer some data, and this is the data the AI will use to learn from. And these data sets are huge… we’re talking billions of points of information.

That’s why AI can seem “alive” when you chat with it, it has a large pool of information to learn from. It’s like a college professor, lots of information has been digested, so it sounds wise. This is important to know, because anyone trying to push the “it’s alive, be afraid” narrative, is maybe pulling your leg. In fact, the opportunities that AI chatbots open up are almost immeasurable… in truth, we haven’t even scratched the surface yet.

For fans of classic movies where two mud-stomper monsters duke it out for dominance, this is a great year. That’s because that same energy is showing up more and more in this space from big names… for instance, Google has a possible ace up their sleeve. An ace called Anthropic, an AI safety and research company that has been working on a language model called Claude to rival ChatGPT. Anthropic scored almost $400 million in funding from the tech titan earlier this year, they are raring to go.

To be sure, OpenAI may have helped make AI chatbots famous with the ChatGPT name, but they aren’t the only heavy on the block. Anthropic has thought about its selling points and is touting Claude as “a safer assistant” designed as less likely to give harmful output. Google must expect big things from Anthropic, because they’re reported to be receiving a 10% stake in the company. Anthropic will use Google Cloud advanced GPU and TPU clusters to train, expand, and implement AI systems.

Of course, not everyone is so impressed… that includes some who criticize Anthropic because their Series B funding came from none other than Sam Bankman-Fried, CEO of FTX, last April. For those folks, never fear. With the funding floodgates being thrown open, the options are beginning to abound… options such as Character.AI. This AI chatbot startup just raised $150 million in funding led by Andreessen Horowitz, placing its valuation at a whopping $1 billion already.

We will absolutely continue to see more AI chatbots everywhere, soon… that includes finance. Remember how we talked about the “environment” where chatbots are trained? Well, Bloomberg just announced that even they are in on the action. Their huge, well-aged set of financial data is perfect for training AI chatbots (although, anything in the works will also access shared language model resources).

According to the announcement, Bloomberg gathered a team and came up with BloombergGPT… and they want it to be your go-to when you think about AI-powered business news. To give you an idea of what BloombergGPT will “know”, it’s training environment (all the data the company will use to teach it with) consists of over 700 billion word fragments (known as tokens). For comparison, that’s 200 billion more than GPT-3, which was released in 2020.

What’s next… AI chatbots in cars? Yes. GM has recently announced they are studying putting them in future models. This would take the services we already use, like Android Auto and others, and kick them up quite a few notches. A conversational AI on the road could make a big difference in driving safety, at least that’s the dream… we’ll see. Come back next week, we’ll have more from the IPO space. See you then!