If you are closely tracking the progress of Open AISam Altman’s company uses neural networks to write original text and create original images with incredible ease and speed.
On the other hand, if you’ve only vaguely paid attention to the company’s progress and the increased traction other so-called “generative” AI companies are suddenly gaining, and want to better understand why, then you’ll benefit from this interview. can do. James Currier, five-time founder, current venture capitalist and co-founder of the company. NFXMore Five years ago, with some of his serial founder friends.
Mr. Currier falls into the group of people who closely track progress. NFX has made a number of investments related to “generative technology,” as he describes it, and it catches the team’s attention every month. In fact, Currier doesn’t think the talk about this new wrinkle in AI is hype, but that the broader startup world is suddenly facing an enormous opportunity for the first time in a long time. not. “Every 14 years there’s a Cambrian explosion like this. There was one on the internet in 1994. There was one of him in 2008 with mobile phones, and another one in 2022. .”
In retrospect, this editor should have asked a better question, but I’m learning here too. Excerpts from chat follow have been edited for length and clarity.you can hear our long conversation here.
TC: There’s a lot of confusion about generative AI, how new it is, or whether it’s just become the latest buzzword.
JC: I think what happened to the AI world in general was the sense that you could have deterministic AI that could help identify the truth about something. For example, is it debris from a manufacturing line? Is it the right meeting to hold? We are using AI to make decisions in the same way that humans make decisions. This is the bulk of AI in the last 10-15 years.
The other set of algorithms in AI are these diffusion algorithms, which aim to look at a huge corpus of content and generate something new out of it, and say, “Here are 10,000 examples.” Can you create a 10,001st example like
Until about a year and a half ago, they were pretty fragile and pretty fragile. [Now] Algorithm improved. But more importantly, the corpus of content we’ve seen has grown simply because of more processing power. What happened is that these algorithms are riding Moore’s Law — [with vastly improved] Storage, bandwidth, speed of computation – and suddenly it could produce something very similar to what humans produce. Meaning that the values are very similar to what humans do. And it’s all been done in the last two years. So it’s not a new idea, but it’s hitting new limits. That’s why everyone looks at this and says, “Wow, this is magic.”
So it was computational power that suddenly changed the game, and wasn’t it part of the previously missing technological infrastructure?
It didn’t change suddenly, it just changed gradually until it reached a point where the quality of that generation made sense to us. The algorithms are very similar. These diffusion algorithms are slightly improved. But really, it’s about processing power. Then, about two years ago, [powerful language model] GPT, on-premises compute, came along, then GPT3. [the AI company Open AI] will do [the calculation] for you in the cloud. The data model was so large that it had to run on its own server.you can’t afford to do that [on your own]And at that point, things really skyrocketed.
we, Company I’ve been doing AI-based generative games, including an “AI Dungeon”, and I believe the majority of all computations in GPT-3 were done through the “AI Dungeon” at some point.
So does “AI Dungeon” require a smaller team than other game makers?
That’s absolutely one of the big advantages. No need to spend a lot of money to store all your data. With a small group, you can create dozens of gaming experiences that all take advantage of it. [In fact] The idea is to add generative AI to older games so that non-player characters can actually say more interesting things than they do now. , for adding AI to an existing game.
So is the big change in quality? Will this technology plateau at some point?
No, it will always improve incrementally. It’s just that the difference in increments will get smaller over time because it’s already pretty good.
But the other big change is that Open AI wasn’t really open.They produced this amazing thing, but it was unpublished and very expensive. Stability AI They said, “Let’s make an open source version of this.” And at that point, the cost had dropped by a factor of 100 for him in just 2-3 months.
They are not derivatives of Open AI.
All of this generation technology is not built solely on Open AI GPT-3 models. it was just the first one. The open source community is currently replicating much of their work and is probably eight months, six months behind in terms of quality. But it’s going to get there. And since the open source version costs 1/3, 5, or 20 times less than open AI, the price competition will be fierce, leading to a proliferation of these models competing with open AI. increase. In the end, you’ll probably end up with 5, 6, 8, or 100.
A unique AI model is built on top of it. So you might have an AI model that actually looks at writing poetry, or an AI model that looks at how to create a visual image of a dog or dog fur, or an AI model that specializes in writing sales emails. Hmm. Take whole layers of these specialized AI models and build them on purpose.in addition Them, you get all generation techniques. How do you get people to use your product? How do you get people to pay for your product? How do you get people to sign in? How do I receive it? How do you create network effects?
who makes money here?
The application layer where people go after distribution and network effects is where you make money.
What about large enterprises that can incorporate this technology into their networks? Wouldn’t it be very difficult for a company that doesn’t have that advantage to come out of nowhere and make money?
I think what you’re looking for is something like Twitch where YouTube could have integrated it into their model, but it wasn’t. Twitch has created a new platform, a new piece of culture and value for investors and founders, but it has been difficult. So there will be great founders who will take advantage of this technology and try to give them an edge. And it will make a seam in the market. And they can build multi-billion dollar companies while big companies do other things.
The New York Times is piece Recently, we’ve featured a handful of creatives who said the generative AI apps they use in their respective fields are tools in their broader toolbox. Are the people here naive? Are they in danger of being replaced by this technology? You’re right, the AI Dungeon team is small. That’s good for the company, but potentially bad for developers who would have approached the game differently.
I think most technology has some kind of discomfort that people have [for example] Robots will replace jobs in car factories. When the Internet came along, many direct mailers were horrified that companies would be able to sell direct mail without using paper-based advertising services.However [after] They’ve embraced digital marketing and digital communication via email, perhaps making a big leap forward in their careers, becoming more productive, faster and more efficient. The same thing happened with online credit cards. Until maybe 2002, we were reluctant to take credit cards online. [this wave in] 2000-2003 had better results.
I wonder what’s going on now. Writers, designers, and architects who think positively and adopt these tools to achieve 2x, 3x, or 5x productivity gains will do incredibly well. I believe that over the next 10 years, the entire world will become more productive. This is a huge opportunity for 90% of people to do more, be more, make more and connect more.
Do you think it was a mistake for Open AI not to do so? [open source] What was it building, given what was born around it?
Leaders will behave differently than followers. I do not understand. I’m not in the company, so I’m not sure. What I do know is that there is a large ecosystem of AI models, and how AI models remain differentiated as they all asymptote towards the same quality and only become price competition. It’s not clear who will win. It looks like Google Cloud and AWS will win.
Open AI can move up or move down. They could become like his AWS themselves or start creating specialized AI to sell to specific industries. I think everyone in this space has a chance to do well if they navigate properly. They’ll have to be smart about it.
NFX’s site has more information. Generative AI By the way, this is worth reading. can be found here.