GPT-4 and the Future of Gen AI in Business: Insights from Sameer Dholakia at Bessemer Venture Partners

Dholakia discusses how more capabilities and shorter gaps between model releases could impact use cases and investments in this field in the future.

Published on Jun 30, 2023

The nascent stages of a new or emergent technology are always a rollercoaster of investment and innovation. This often creates exciting and competitive bouts between businesses over a short time. Newcomers on the scene deliver combo moves and well-established incumbents follow up with counterpunches.

Winners are difficult to predict before the dust settles but one thing is generally certain: if the technology is powerful enough, the business world often thinks of it in terms of life before and after its arrival.

Generative AI is one such technology that’s newer but is already having a profound effect on business. The viability of its functionality and use cases is creating a sometimes combative, sometimes cooperative relationship between newcomers and incumbents pushing themselves toward greater market influence. It’s also led to a white-hot investment scene, where funding to this point in 2023 has already surpassed the $1.5 billion raised across all of 2020.

The recent release of OpenAI’s highly anticipated GPT-4 model put all of this into perspective even further. Its multimodal functionality is very impressive and it opens the door for even more creative use cases that businesses, like Microsoft, Duolingo and others, are already taking advantage of.

However the technology won’t stop evolving there. GPT-5 and other models and use cases will emerge to the tune of more investment funds and competition.

Sameer Dholakia

I was curious about how GPT-4's arrival is seen within the context of this competitive investment and product development landscape we’re in now. So I asked Sameer Dholakia, a partner at Bessemer Venture Partners and former CEO of SendGrid (which he led to an IPO) what his thoughts were on the subject.

In our conversation, he examined the current and projected landscape of generative AI investment and development through the lens of GPT-4’s release and its implications. He commented on everything from the widening gaps in functionality between major model releases like GPT-3 and -4 to the questions that venture capital firms have to ask themselves when examining the direction of this wild market.

Read on to hear his expert perspective.

What impact do you expect GPT-4 to have in terms of underlying generative AI functionality?

What is most astonishing to me is the leap forward in terms of it being multimodal. It’s more human-like in its ability to see an image of something with text, process both, and reason about what that means. That really is extraordinary.

I think the multimodal leap forward is actually quite important. It sets a new baseline of what is still to come. It's now up to the innovators, entrepreneurs and the rest of the world to figure out use cases and ask, “So now we can do this multimodal thing. What problems can now be solved that couldn’t before?” From a VC perspective that's the kind of stuff that we're constantly looking for: “What are the wider applications of this particular innovation?”

So you could imagine in the world of healthcare, for example, using generative AI to examine an X-ray with the logs of previous visits. You can determine what was ascertained in previous medical tests with text and images. That’s just one example of many where the multimodal capabilities of GPT-4 could be really interesting.

What about in terms of business impact?

The thing that is most exciting to me about GPT-4 is what it portends for the future of GPT-5, -6, -7, and so on based on looking at the delta in output between GPT-2, -3 and -4. When you submitted something to GPT-2, I felt like the answer was maybe the equivalent of a middle schooler. GPT-3 was high school and -4 is collegiate (even PhD) level output. It just keeps getting smarter. The improvements are steepening and those cycle times are getting shorter.

That's a remarkable thing. We've seen that happen already in the history of technology. Take Moore's Law with microprocessors. We're seeing that again and I’m sure somebody will develop a formula and coin a phrase that will be the equivalent of Moore's Law for Generative AI models — likely noting the extraordinary expansion in model parameters in each successive generation.

GPT-2 had 1.5 billion parameters. GPT-3 had 175 billion parameters. GPT-4 was rumored to have 100 trillion parameters but OpenAI CEO denied that. However, given the massive jump in parameters from model to model, GPT-4 is likely to have far more than its predecessor.

So the business impact of GPT-4 is continuing the tidal wave that’s already coming. It's going to make it so much easier for every software company on the planet to embed these capabilities and for generative AI technology to be used by billions and billions of humans. ChatGPT hitting 100 million users is just the tip of the iceberg.

I can’t predict whether GPT-4 or -5 will be the tipping point. But there's no question in my mind that that's where we're headed. And we can see it coming with announcements from Microsoft and Google about them embedding this technology in their products (e.g. MS Office and Google Apps). I think the business impact of these releases are profound because it's going to be so much easier for companies and providers like them to put the technology in the hands of billions of humans that are already using tools like Word or Docs, Powerpoint or Slides.

One often hears that “we’re in the early innings of the GenAI revolution.” If I were to liken it to previous platform shifts that I’ve seen, I’d say we’re in about 1995 of the internet (Netscape’s browser making the web more accessible, and entrepreneurs like Jeff Bezos figuring out how to capitalize on the new capabilities). Or, if you prefer a more modern analogy, we’re in the summer of 2008 of the mobile revolution (when Apple introduced the App Store and enabled developers to build ios-native apps.) We couldn’t imagine then that entrepreneurs like Jack Dorsey, Evan Williams and Biz Stone would use it to catapult Twitter or that Travis Kalanick would use it to create new services like Uber; these developments would have been impossible without it.

What makes me most excited about this period that we’re currently living in is that we are about to see the creative brilliance of the next generation of entrepreneurs unleashed as they grab hold of this new platform and bring new solutions to market. And they change the way we live and work in the process.

Where else do you see generative AI technology making major impacts as the technology advances?

I really liked Meghan's (Jasper CMO Meghan Keaney Anderson) presentation at the Gen AI conference. I think it was one of the best examples of how generative AI is going to be a productivity multiplier, like a copilot for a content marketer to make them five times more productive. Right now, there’s so much time over-indexed to writing and content generation. Using a generative AI tool lets people redistribute time to higher value stages of the writing process (like ideation, research, and editing), thereby creating more work satisfaction and engagement.

There have been so many times where technology has come along that made people say, “This is going to impact so many jobs. What does that all mean?” In my lifetime, some jobs do end up being impacted — but what I've seen more often than not is that people just work on higher value added activities, because technology and automation removes the mundane, and allows people to work on more interesting stuff.

A slide from Jasper CMO Meghan Keaney Anderson's Gen AI presentation

With the time gap between major models being shorter but the advances in their capabilities growing, what might this speed of innovation mean for future investment outlooks?

I think it just heightens the focus around differentiation and the risk profile. We’re at a state where as the improvements accelerate, it's easier for incumbents to embed the capabilities themselves.

The debate we've been having a lot from a VC’s perspective is, “There's no question that this is a transformative new technology, but who wins in that? Where does the value accrete?” Then we try to draw some analogies and make some predictions. Will it be the incumbents, who already have customer relationships with businesses, that just embed this new capability in their own software? Or will it be net new disruptors (call them the AI-natives) who show up with a new capability leveraging generative AI; organizations that are moving fast and not waiting for tomorrow. Historically, you bet on the innovators.  But some would argue that, like the cloud computing wars, scale matters. So perhaps (at least at the infrastructure layer) the incumbents win. No one has a crystal ball and I suspect the answer will indeed differ by category and layer of the tech stack.  

Another thing to consider is, given how quickly this capability is maturing and how easy it is to access with an API call, the barrier to adoption is so much lower than it historically is when you have a platform shift.

Look at the shift that companies went through when moving from on-premise software to using the cloud. If you were an on-premise software company, you had to change everything — from your business model to your pricing to your software development process to your infrastructure — to become a cloud-native SaaS company. The barrier to entry of that platform shift was massive. Therefore, you bet on the Salesforces of the world (who did end up disrupting Siebel) — but this platform shift is different. Adapting here is just a matter of calling an API or putting an open source LLM model into a tech stack. It doesn't have the same disruptive hurdle that these other platform shifts have historically had. As a venture community, we have to figure out what all that means in terms of value capture.

What you're seeing today around generative AI is a massive inflow of bets by the venture community on this new platform shift. Some are doing it at the foundation layer, some are doing it at the application layer and some are doing it with companies that are doing both (that have their own model and their own application.) There's no obvious right, or mutually exclusive answer. Or at least we won't know what it is for another until for another five to 10 years, but we'll see!

Meet The Author:

Alton Zenon III

Jasper Content Marketing Manager

Want more of The Prompt? Sign up for the newsletter and never miss a story.

Receive a weekly recap of breaking news, insights, and interviews with experts in the world of generative AI.

Ready to create better content with AI?

Discover how 100,000+ copywriters and businesses are using AI content.
Works in 30+ languages
2 minute signup
Rated 4.8/5 stars in 10k+ reviews