The GPT Store is LIVE! (But Does Anyone Care?)

Today on the AI breakdown, we're looking at the just released OpenAI GPT store. Before that, on the brief Valve reevaluates its policy towards AI powered video games. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to 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. We kick off the day with an update from Valve. Valve are the makers of Steam, which is of course the platform that tons and tons of independent games are distributed through. Now the company used to have a policy that game developers could not use AI or generative AI specifically in the creation or execution of their games. However, when they put that policy in place last summer, they said it was to give themselves time to figure out how this was all going to work and update their policies. Well, update their policies they have, they write. Today, after spending the last few months learning more about this space and talking with game developers, we're making changes to how we handle games that use AI technology. This will enable us to release the vast majority of games that use it. Basically what will happen is that developers will need to disclose when they're using AI. They'll have to say when their games contain pre generated content, and they'll have to promise that it's not illegal or infringing on other people's copyright. They'll also have to specify if their game has AI content that's generated live during the gameplay itself. And if that is happening, they'll have to explain the safety measures they put in place to make sure that that AI doesn't go rogue and start generating content that would otherwise be illegal or infringing. Games that use AI will be identified as doing so, and players will have new options to report illegal AI generated content that comes up in games that they're using now. The Verge sums up the tension in the game industry about all of this as such, while some developers have been keen to integrate the new technology into their games and production processes, the wider industry is split on the use of generative AI. On the one hand, several studios have talked about using AI to help with game testing, early concepting, or helping with expensive parts of the game development process like voice acting, recording sessions. But others fear that AI could be used to cheaply replace existing artists and other creatives and are pushing back against companies who publish AI generated assets. Just a couple days ago we talked about the whole dust up with Wizards of the coast putting out AI generated marketing materials after having committed to not having AI in their games. And this is the type of thing that we're just going to see a lot more of. However, if I had to guess which way the winds will ultimately flow, I think it's more like this and towards broader acceptance of using generative AI and harnessing the benefits of generative AI than the other direction. But I could be wrong, and there's certainly some evidence that broad public opinion is not in favor of artificial intelligence. Daniel Collison and the AI Policy Institute have released their latest poll and it shows a continued and growing interest among the American public for having more guardrails around artificial intelligence. Daniel tweets A majority of Americans support comprehensive AI legislation in 2024, 56% support versus 22% against. Those respondents also support consideration of legal liability for criminal usage of models at 90%, restrictions on deepfakes at 82%, capability testing and certification for models at 91%, and limits on the capabilities of models at 83%. Overall, 56% of Americans polled agree it would be a good thing if AI progress was stopped or significantly slowed, with just 27% disagreeing. Now, these numbers are fairly stark, but I wonder to what extent it reflects A the scariness of media headlines around AI that people have experienced now for a year, and b, just a general sense that there should be some guardrails and a more broad lack of trust in big tech. Well, it certainly is a growing political issue. Yesterday, the Senate held a hearing on artificial intelligence and the future of journalism. According to the expert witnesses who were invited to testify, as Time put it, AI poses a grave threat to journalism. Now, it is notable that the people who were invited to testify were themselves basically just media executives. It was people like Roger lynch, the CEO of Conde Nast, who argued that generative AI tools had been built with, quote, stolen goods. Lynch called for, quote, congressional intervention to make AI developers pay publishers for their content. He said, quote, the amount of time it would take to litigate, appeal, go back to the court's appeal, maybe ultimately make it to the Supreme Court to settle. Between now and then, there'll be many, many media companies that would go out of business. Now, of course, this is very heated rhetoric right now from a cohort of people who see yet another threat to their business model. But ultimately, this battle is one that is going to be fought in courts almost no matter what rules Congress passes. Where there does seem to be broad agreement is that misinformation is a big threat of AI right now. A new Davos report has just come out, in fact, that suggests that AI powered misinformation is the world's biggest short term threat. The World Economic Forum published its latest Global Risks report, which is published ahead of its annual event in Davos. When asked what the most severe threat was over the next two years, misinformation and disinformation was the most cited threat. Given that Davos attendees will include OpenAI CEO Sam Altman, Microsoft CEO Satya Nadella, Meta chief AI scientist Yann LeCun. You can expect to hear a lot more about AI from that global elite gathering in the next couple of weeks. Now, one thing that I think is notable is that on almost any day that you hear some congressional hearing about the dangers of AI or some Davos report on how misinformation is so scary, you're also probably going to find a press release from some military establishment. Very often the US Military establishment, talking about how they're taking advantage of AI for advanced military capabilities. For example, I just came across this blog post US Department of Defense Increases AI Capacity through Strategy and Alignment the Defense Department is making leaps when it comes to fielding cutting edge technologies, a senior Pentagon policy official said today. Michael C. Horowitz, Deputy Assistant Secretary of Defense for Force Development and Emerging Capabilities, said key organizational and strategy updates have resulted in DoD's improved ability to effectively field new tactics and technologies, especially when it comes to artificial intelligence. Basically, I think that the lesson here is that for all of the jawboning that there might be, this technology is being implemented in numerous ways across many fields. Every minute that those gums are flapping, I'll continue to watch for the most important advancements in that, but for now that is going to do it for the AI breakdown brief. I'll be back soon for the main AI breakdown coming off of November's OpenAI Dev Day. Easily the most exciting part of the announcement for most people was that the company had created a new way for people to create what they were calling custom GPTs, which were effectively specific purpose versions of ChatGPT that could come with preset instructions and even custom actions that went outside of ChatGPT GPT. And alongside that there was also a promised GPT Store. Now, as soon as the GPT store and GPTs were announced, many people started making the comparison to Apple and the App Store. Initially, the GPT Store was set to go live in December, but then based on, oh, I don't know, the extreme leadership chaos that ensued that got pushed back. But last week they sent out an email to creators of GPTs that the GPT Store was going live this week and sure enough, yesterday it actually happened, the blog reads. We're launching the GPT Store to help you find useful and popular custom versions of ChatGPT. Couple of the key things from this announcement post first of all, in the two months since announcing GPTs, over 3 million had been created. Second of all, there was nothing in this announcement that explained how GPTs would actually be monetized again if the natural comparison to make was the App Store. There isn't necessarily the same sort of pay for a particular GPT kind of approach that worked for apps, OpenAI writes in Q1 we will launch a GPT Builder revenue program. As a first step, us builders will be paid based on user engagement with their GPTs. We'll provide details on the criteria for payments as we get closer. Now lastly, as a little bonus to the announcement, they also announced a new ChatGPT team plan, which is sort of like in between GPT plus for individuals and ChatGPT Enterprise customers that use the team feature get access to a private section of the GPT Store so that people can share GPTs that they build for their particular workplace with just their other teammates. I think this makes a lot of sense given how much of the early usage of these GPTs I believe will be just for people simplifying their own workflows by not having to enter in the same prompt instructions over and over for use cases that they come back to week over week or even day over day. Now, in addition to that dedicated space for private GPTs that are shareable across your team, team users also get GPT4 with a 32k context, an admin console, a guarantee of never training on business data or conversations, and the price is just a little bit more than the plus subscription. It's $25 per user per month if billed annually, or $30 per user per month if billed monthly. But back to the GPT Store itself, let's take a look at how it's actually structured. First of all, right at the top you have a search field, but really one of the things that people were looking for, and certainly I was looking for, was a better way of navigating what is an extremely dense set of prospective GPTs. Remember they said that 3 million have been made, and while all of those aren't necessarily shared publicly, that still means you're talking about probably hundreds of thousands of public GPTs, which could make it very difficult to find exactly what one is looking for. So how are they going about solving that browsing challenge. Well, first of all, they have a curated top picks from the week that they call a featured section right out of the gate that includes an All Trails GPT for finding hikes, a code tutor GPT from Khan Academy, a books GPT and an AI research assistant called Consensys. Under the top pick section. They also include trending GPTs, which is just a measure of what's being used most interestingly right out of the gate. A lot of them are research assistants. I don't know if that's because of a concentration of students and researchers being a big part of the user base or because they're a really accessible thing to try, but it's still pretty interesting. Other notable trending GPTs include Grimoire, which is a coding GPT up at number four, Canva, which is a design GPT obviously at number eight, and both a video generation GPT and an image generation GPT. There's also 12 GPTs that are created by the ChatGPT team itself. These include things, you know, like Dall E, as well as an instantiation of their advanced data analysis which used to be called Code interpreter, but they also have things like Coloring Book Hero, a GPT that explains games, and a creative writing coach. From there they've also organized things into common categories. The categories they feature include Dall E, so things that are in and around image creation and generation, writing related GPTs, productivity related GPTs, research and analysis, related GPTs, programming, education and lifestyle. With of course lifestyle being a catch all for hobbies, travel, workout style, food, et cetera. Now let's take a quick look at the Canva GPT because I think it does a pretty good job of showing Both what these GPTs are useful for, but also their limits. So Canva effortlessly design anything presentations, logos, social media posts and more. Canva is of course a design suite that lives on the web that is a competitor to things like Photoshop. Lets write Please design a YouTube thumbnail for my video about the new ChatGPT GPT store. It says Canva wants to talk to chatgptplugin.canva.com we'll allow it and what it provides are two different options. They say hello, here are two creative YouTube thumbnail designs for your video about the new ChatGPT GPT store. Feel free to choose the one that best suits your vision. Remember, you can further customize the design in Canva to match your specific color or theme preferences. So let's say that I like option two a little bit better. Well then what's going to happen then is it's going to bring that up in Canva, where of course I can customize the image, the text, the colors, et cetera, et cetera, et cetera. And so effectively what I've done here is I've used a natural language input via ChatGPT to prompt Canva to start and create kind of a template or a working model for what I want. In other words, instead of going to canva.com and searching for YouTube thumbnail templates, I've described what I wanted with natural language in ChatGPT. But that's really the end of the difference of experience. It is effectively a different UI option for how to get started with the Canva creation process, which may be a a preferable user experience for some, and b it may be a faster user experience for some. What it's not is some total replacement that changes everything. And I think that having some amount of limited expectations for what GPTs, at least right now, can do is certainly going to behoove users who are otherwise potentially going to be disappointed. And there is frankly a lot of skepticism out there. Sim4Satoshi writes GPT store is absolute hot garbage. I'm willing to bet the retention metrics at the end of this month for GPT Store will be abysmal. Anyone in OpenAI want to take this bet? Yana Wellander, the CEO at Craftful writes, Wild GPT Store just launched and there's already nine imposters to Craftful GPT. Many people have pointed out that when it comes to copying GPTs, it's really easy because in many cases people can just ask what is your exact prompt? And those GPTs will explain exactly what their prompt is. So much so that people are now taking to Twitter or X and explaining how people can try to build in custom instructions to not allow their GPTs to do that. Alex Utopia wrote a pretty thorough skepticism of the GPT Store that I think is worth consideration, he wrote. I'm not entirely sure what I expected, but what we've received from OpenAI is rather underwhelming. The GPT Store is being hindered by issues related to exploits and the business model. The problems he identifies are one disconnected content curation. The GPT store presents seven categories, each featuring 12 GPTs on the front page, and it is highly curated. The section aligns with what you would expect from an AI developer, coding assistants, AI research paper search engines, and more coding helpers. Apparently OpenAI has no marketing team or anyone capable of thinking like a regular person where are the party games, the dating wingman apps, and the resurrected historical figures? In short, the store isn't much fun, although I know these GPTs are out there somewhere. 2 he writes. Who is this for, exactly? Even if there were more interesting items in the storefront, the GPTs are no good to anyone without a ChatGPT subscription. And people who are already customers are most likely familiar with all the functions and able to build their own custom GPT. 3. Scraping and exploits the only way I see to make a truly useful GPT is to feed it with a lot of data that only you have access to. However, it is still possible to extract all the documents and prompts from GPTs. With clever prompting, any skilled user can simply scrape your data and build a copy of it. In conclusion, I'm uncertain about the utility of this it certainly doesn't seem like OpenAI's App Store moment now. That said, not everyone is so skeptical. Vox's article is titled There are too many chatbots. Will OpenAI's new chatbot store finally make AI useful? ZDNIT writes, GPT Store is OpenAI's most compelling reason to become a ChatGPT subscriber yet. And even more than that, a wave of people are starting to give their tips and advice on how to make GPTs better. Rowan Chung from the Rundown writes, OpenAI's GPT store is here. Most GPTs will be useless, but if you want to make a useful GPT rank high and make money, you need to use actions. He then proceeds to give a tutorial on how to add actions to custom GPTs. Others are creating ways for GPT builders to add things like analytics, or to collect emails, or even to monetize directly. My sense is that right now the right approach, even if one finds themselves skeptical, is to have at least a little openness to the possibility that this is very early and that naturally the first crop of things that are created in something like a GPT store are going to be wildly all over the place in terms of both their utility and excitement. Ultimately, we will see from usage if any of these things are actually sticky, and I think that's going to be pretty interesting to watch. No matter what, it exists now, you can go play with it, you can go check it out. So have fun gpting, and until next time, peace.