The AI Daily Brief: Artificial Intelligence News and Analysis - 5 Predictions for the GPT Store Coming Next Week

Episode Date: January 5, 2024

OpenAI announced the GPT Store is finally launching next week. NLW gives 5 predictions on how it will play out. Before that on the Brief: rumors of AI-powered Siri updates from Apple and a leaked Bard Advanced powered by Google Gemini Ultra. ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, five predictions for the GPT store, which is coming next week. Before that on the brief, rumors from Apple and Google about forthcoming AI efforts. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Could a breakdown network for more information about our YouTube, our Discord, and our newsletter. Welcome back to the AI breakdown brief, all the AI headline news you need in around five minutes. Today we have quite a rumor-filled brief with some pretty interesting things, and we start off on Google Bard, where the company appears to be on track to release a paid Bard-ad-ad-advanced version that is powered by Gemini Ultra.

Now, of course, you'll remember that Gemini Ultra is Google's big GPT-4 competitor. In December, Google formally announced Gemini and discussed three versions, one that was small enough to run on a phone, one that would be available right away called Pro, which was a sort of GPT3.5 level model, and Gemini Ultra, which purported to actually beat GPT4 in some tests. Now, the problem, of course, with that announcement was that Gemini Ultra wasn't available yet, so no one could really tell if it was actually true. Well, now we've got a report from a developer who has found in the code base some indications about what might be coming down the pipeline from Bard.

The developer in question is named Dylan Russel, and yesterday wrote a thread, Let's Talk about Bard. Dylan writes, while extensions in Bard launched in September, I shared with 9 to 5 Google a first look at the feature back in July. It's been a couple of months since I shared anything about it, so what is Google working on? There are some exciting new features coming in Bard. Important note, these are all unreleased features, they may change and they may never land

at all. One, Motoko. There isn't much about Motoko yet. However, there is just enough for me to tell you that Google will allow you to create bots. I can't say what they mean by bots exactly, if they will be shareable or anything, but yes, Matoko is the codename for bots. Number two, Bard advanced with Google 1. Google will allow you to get three months of Bard advanced on them through Google 1.

Bart Advanced will use Gemini Ultra. Three, Gallery. While the feature is not yet finished, gallery should allow you to explore different topics to see what you can do with Bard. In other words, when this page will be public, it's likely that it will display example prompts about various topics. Four, tasks. The tasks page is the page I know the least about. From what I could put together, though, it appears that it may be used to manage long-running tasks, such as generating an image. Number five, sharing background and foreground. This one is pretty explicit. Although, while my screen touch are showing what foreground looks like, it doesn't show background. Number six, power-up. This feature, once it's working, will allow you to improve your prompts by theoretically

running them through an AI to expand them. I don't have any example of the feature working, unfortunately. There you have it. BARD is a seriously interesting AI. It's amazing that it's free. All right, so a couple takeaways from me. First of all, I don't think it's necessarily a surprise that BART is getting an advanced version that's powered by Gemini Ultra, and it's not really necessarily a surprise that that would be a paid model. Given that, for example, to get GPT4, you have to pay for ChatGBT Plus. However, it does seem like cost is one area where a giant like Google could try to compete

to try to win some market share back from ChatGPT. Ultimately, it still all comes down to, does Gemini Ultra outperform GPT for? The power-up I also think is pretty interesting. This, I think, points to a future in which prompt engineering is less emphasized, and AI helping you perfect your prompt to get the results you want is more of the process and workflow of how things happen. Now, of course, Dylan wasn't able to confirm that this is exactly how it works, but should it be what he thinks it is, that will certainly shape how this new set of people interacts

with AI. On Motoko, this bot creator, I don't know what to think yet. Is this their version of the GPT store, which is, of course, the subject of our main episode today? I don't know, but it's certainly intriguing. Now, Google wasn't the only big tech AI competitor with some interesting reported news. According to a post on the Korean social media site, Navar, from a user who has previously posted correct Apple rumors, Apple is getting ready to premiere a generative AI-powered version of Siri at the next worldwide developer conference. Wright's Mac rumors, according to developer sources speaking to the operator of tech news aggregator

account, Yoix 112 on Navar, Apple has recently made progress with integrated. generating generative AI into Siri using its Ajax-based model that Bloomberg's Mark Germann first mentioned in July. The new version of Siri apparently touts natural conversation capabilities as well as increased user personalization. The new features are believed to be available across devices, suggesting that the new version of Siri will retain conversation information from one device to another, and Apple is purportedly working on linkages for the new version of Siri to connect to various external services, likely via an API. Finally, the report suggests that some of Apple's new AI features may differ based on subscription service status, but it is unclear

what shape this could take. Now, to some extent, I think that many people will take this as confirmation of what seems fairly obvious. Siri is 100% the low-hanging fruit when it comes to what Apple could do with generative AI. It is desperately in need of a major update and shift, but is already so central to the Apple OS experience that it could become the first, quote-unquote, AI assistant that many people interact with. Now, moving over to the land of OpenAI for a moment, one of the things that was notable about the resolution of Sam Altman's firing and then rehiring was that the final board composition wasn't determined. Instead, where the compromise was met was that Altman and ally Greg Brockman would return to the company but not to the OpenAI board and that the initial board would be

made up of Cora's Adam DeAngelo, the sole remaining member from the original board, former Treasury Secretary and Harvard President Larry Summers, and as board chairman, former META employee and Salesforce CEO Brett Taylor. Well, according to the information, OpenAI has been discussing board seats with a variety of other AI notables. Some of the candidates include Scale AI CEO and co-founder Alexander Wang, as well as former GitHub CEO and investor Nat Friedman. Now reporting during the whole crisis held that these were also two of the people that the board approached to potentially be the interim CEO of OpenAI in the wake of Altman's firing. If I remember correctly, both of them turned OpenAI down. Writes the information, it's not clear who else the board has approached about

joining its ranks. But the talks, which are still in the early stages, indicate that the board is seeking to add representatives from companies that partner with the chat GPT maker such as scale or that work with other AI startups. There is a lot that's still to be determined about the exact composition of OpenAI's board. First of all, it's not clear how many seats they're actually trying to fill. The board at its biggest had nine members, and so that might be what they're going for, but that hasn't been articulated anywhere. Secondly, the type of person that they're looking for is also unclear. Part of the issue with the previous board is that there wasn't really a balance between the business consideration and the nonprofit or mission-oriented considerations,

and so are they going to try to redress that balance? That's something to watch and see. One of the challenges for them is that many of the people who are qualified have conflicts of interest with other companies. That's what led Reid Hoffman to leave, given his relationship with potentially competitive companies like inflection, and that's also what led Shabanzillas to leave. Now, when it comes to Microsoft's presence on the board, as part of the agreement, they will not have a full board seat, in other words, a full voting board seat, but they will

have at least one board observer. The information suggests that the leading candidates are deputy CTO, Lila Tretakov, and VP of Partnerships, Deanna Templeton. Now, the other little bit of interesting OpenAI news has to do with their negotiations with publishers. Of course, the biggest news from the holiday break was that the New York Times had sued OpenAI and its partner Microsoft, and part of the reason for that was that they couldn't come to agreement in private negotiations around licensing the New York Times content. The information once again published a piece yesterday called OpenAI offers publishers as little as a million dollars a year, writes the publication, OpenAI has offered some media firms as little as between 1 million and 5 million annually to license their news articles for use in training its large language models, according to two executives who have recently negotiated with the company. That's a tiny amount even for small publishers which could make it difficult for Open AI to strike deals. Now, of course, we've simultaneously got reports that Apple is willing to spend up to $50 million a year on similar deals, although we don't know across how many publishers that split. I think it's interesting that the information is taking such a hard line that this is clearly too little when we're in total. totally uncharted territory when it comes to what these articles are being licensed for.

AI training is categorically unlike other types of licensing uses, and even if that doesn't mean that ultimately courts will find that AI training is covered by fair use, as obviously all of these big labs are arguing, and maybe that the economics look different than some of those other licensing use cases. In either case, one of the things that I'm interested to see with all these court cases is how OpenAI frames its attempts to pursue these licensing deals, in contrast with its argument that its training on copyrighted material should be covered by fair use. So, as I said at the beginning, lots of interesting things going on behind the scenes. That is going to do it for today's AI breakdown brief.

Next up, the main AI breakdown. Welcome back to the AI breakdown. Yesterday, I, with so many other people, got an email from OpenAI with one of the more anticipated things ever since back at their OpenAI Dev Day in November. The email read, Dear GPT Builder, we want to let you know that we will launch the GPT store next week. If you're interested in sharing your GPT in the store, you'll need to review our updated usage policy and GPT brand guidelines to ensure your GPT is compliant, verify your builder profile,

publish your GPT as public. GPTs with anyone with a link selected will not be shown in the store. Thank you for investing time to build a GPT. Now, of course, GPTs are what OpenAI calls their customized versions of, chat GPT, which they released a builder for back again at Dev Day. Effectively, they are custom instructions that you can program for any sort of purpose that you want. They can be really simple, just the instantiation of a particular use case, such as something where you've told ChatGPT that it's an excellent copywriter, and you want it to always write in marketing copy in bullet

form. Another capacity, though, of the GPT builder is that you can have it reference specific information. You can give it PDFs, for example, that you wanted to pull from. Sam Altman did that in his demo, sharing a blog post that he had written previously as a way to build an advice GPT for startup founders. You can also have GPTs trigger custom actions that happen outside of chat GPT, and that's, of course, where some of the more advanced use cases come from. Now, so far, what I've seen GPTs useful for is for people to standardize workflows that they use over and over again. For example, some of the GPs I've created are an AI content creator, which helps me come up with ideas for AI-related content,

an AI tutorial muse, which helps me come up with different tutorials for the AI education beta that's going on right now, a job description copywriter which is customized for my particular needs, for people that I'm bringing on the team to help with that AI education beta, and then more recently a one-pageer business planner that was designed as part of one of those tutorials for that beta audience. In other words, none of these are explicitly designed for other people to use, other than this one-pageer business planner,

which is part of a larger educational experience. Instead, they're just things that I needed for myself, and the time it took to create a custom GPT was a lot less than the time it would take to reenter a set of prompt instructions every single time. And yet, if those sort of individual use cases are what I've seen predominantly so far, many people wonder if the presence of a store will open up a new set of opportunities. Certainly there are no shortage of GPTs out there. Numerous sites like SEO.a.I rank them based on various criteria.

SEO.AI, for example, uses referring domains, which point back at those GPTs. They have Blog Wizard at the top, Grimwar a coding GPT, up at number seven, and a lot of writing and design-related GPTs that are in the top 25 or so. What plugin.aI is another site for finding GPTs that also has not only a ranking, but a categorization system that includes categories like academic research, coding, text to speech, learning, investment, and trading, and more. So what I thought I'd do today is give you five predictions for how I see the GPT store playing out. These are, of course, totally subjective in just my opinion, but if nothing else, they should give you some food for thought.

Number one, discoverability will be more important than monetization. So what do I mean by that? First of all, what do I mean by discoverability? If you look at sites like this one, what plugin.aI, is very clear that one of the challenges of getting utility from GPTs is just finding one that might actually suit your purpose. Right now, people are relying on Googling to find third-party sites like what plugin. com to have to go try to search, hoping that the keyword that they're thinking about matches how the site has tagged a particular GPT, or otherwise clicking on a category and hoping to find something that fits within it. That is a super suboptimal experience, and suboptimal experiences like that lead to declines in usage.

It creates a barrier to entry for people to find utility. So when it comes to the value proposition of having a GPT store, the thing that I'm most interested in in some ways is whatever system they put in place for people to actually find the GPTs that might be relevant for them. This will presumably be part of the GPT store experience. I can't just imagine it's going to be an endless scrolling list of GPs with the particular price next to them, although I guess that's certainly possible. And so in some ways this prediction is actually saying two different things. One, it's that right now I see the biggest problem not as monetization, because I think people are incentivized to create GPTs that are really useful for themselves already, but

discoverability so that the efforts of the top part of the bell curve of people who who actually want to create these GPs, their work can actually provide value to the long tail because they can now find those GPTs. Now the second part of this is perhaps a little bit of an inherent skepticism around how monetization will work. I think people are still just at the workflow experimentation phase, where unless there is some hyper-specific functionality that it feels very hard to imitate on one's own, I see it is fairly difficult for people to get out their wallets and actually buy these things.

Now, right now we don't have any idea about how Open AI is going to suggest they be monetized, what prices they're going to suggest putting around them, and I think that that could have a major impact on things. But overall, my bet here is that initially, the GPT store will increase usage, but not because they create a monetary incentive around them, but simply because they create a better architecture for discovering the ones that are most relevant to any given user.

Now, staying on the theme of monetization for a moment, though, my second prediction is that the tools that will be most easily monetized are coding and no-code development tools. These are going to be the more comparatively complex GPs that involve external actions, and which take processes which seem difficult or inaccessible now, such as building an application, and turn them into a natural language-mediated process.

I think, in other words, that people are more likely to pay for a thing that allows them to do something that they weren't able to do at all before, as opposed to paying for a thing that just helps them do a thing that they can already do, but better. I think that the gap from zero to one versus one to two, in other words, is a very significant one. If you look back through this list, there are some notable GBTs that are coding assistance, Grimois being maybe the best known. If I was trying to monetize, that is certainly where I would be looking. Number three, a couple of predictions around who are likely to be the earliest adopters and the first power users.

One is a fairly safe guess, I think, in students, given that students are almost always early adopters when it comes to any type of new technology like this. But I can definitely see people finding study companions and topic-specific tutors or research assistants for a particular discipline, a really useful and effective use case right away, and a small price being worth it to shortcut time when it comes to that type of use. Now, the second category that I think are going to be early adopters are digital marketers. The reason for that is fairly simple. Digital marketers spend a lot of time working in the realm of words. They write ad copy. They write social media copy.

They plan social media posts and content calendars. All of these are things that LLMs can be really useful with. And again, having a shortcut of a GPT that's perhaps trained on best practices or has reference examples, And so it's going to be punched up when it comes to copy, relative to just the custom instructions that you might give chat GPT, I can see being a really valuable addition to the digital marketer's toolkit. And once again, I think validating this is the number of tools that relate to digital marketing in this SEO.aI top 100. We've got SEO, de jargonizer, blog expert, storyteller, chat CEO, viral hooks generator, search quality GPT, SEO mentor, article GPT, and more. Fourth prediction is that GPTs will be much more relevant for professional than personal use. What I see GPTs as is as the embodiment and consolidator of professional workflows.

In other words, GPTs are really useful for tasks that a professional has to do over and over again, where they don't want to have to copy-paste a similar set of instructions every single time, or even worse, re-imagined them each and every time they have to perform that task. In many ways, your portfolio of GPTs becomes like a list of common workflows that you use day in and day out, and I think that that's going to be much more valuable in the individual professionals use case than it is for some of the more personal uses like recipe finders. That's not to say that we won't see lots of experiments with random things like recipes and games, I think that we will, but I think that the real utility is going to come from professionals

who are using GPs around particular workflows. Related to that is my fifth prediction that GPs will in fact help some, some roles or professions actually better understand how to use chat GPT because all of a sudden things will be framed in the context of those workflows. Think about it this way. Imagine that you are a chat GPT power user working in a digital marketing department and trying to help your boss understand why chat GPT is useful. Instead of sending them chat GPT, a blank open cursor screen, and telling them to play around with it and try writing some copy, you send them to the copy editor GPT and suggest that they copy paste in some text and see how the GPT improves it.

Basically, I think that not only will professionals be the major users of GPTs and the GPT store, but the fact of GPT and the GPT store's existence will make more professionals users of chat GPT in general. Last prediction, sort of a bonus, and sort of a catch-all, frankly, inevitably even if our expectations of how much monetization there's going to be are way off, the simple presence of having a store will inevitably unlock more creativity. People are going to try to make money because they can, even if it doesn't work. And there's going to be some period before things are figured out, where there will be a

Wild West Cambrian explosion, choose your metaphor of incredible creativity and really interesting and weird and off-kilter use cases, some of which will probably stick in ways that we wouldn't have imagined. I, for one, I'm super excited to see how that plays out, but for now, that is going to do it for these five or six predictions. for the GPT store coming next week. Appreciate you guys listening or watching as always. Until next time, peace.