These Are the Most Used AI Applications
13:52 Share

Today on the AI Breakdown, we're looking at a new report about which AI tools are the most used. Before that on the brief, AI is showing up in company performance, but Wall Street is getting nervous about the costs. 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 newsletter, and our Discord.

Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. Today we kick off with another report from the world of markets and their relationship with AI. Microsoft has seen its highest profit growth in more than two years, with its net income rising last quarter 33%. Said Microsoft CEO Satya Nadella, a growing body of evidence makes clear the role AI will play in transforming work.

Now, Microsoft is interesting because it has very clearly showed that there is a connection between investment in AI and increased profits. And specifically, the connection point for Microsoft, at least, is the relationship between AI offerings and their cloud business. Microsoft's Azure business grew 30%, which beat analysts' estimates of 27%. Microsoft said explicitly that six percentage points of that growth came from AI demand, which represented double the amount AI contributed to Azure growth in the previous quarter.

Now, Google also saw a 13% revenue growth last quarter, but the rebound in its ad sales business was short of analyst estimates, leading the stock to actually fall. In fact, the narrative is less about AI right now from a revenue perspective and more from a cost perspective.

Reuters writes Microsoft touts AI strength but shares dip as market digest costs. Basically, in both the case of Alphabet and Microsoft, part of their earnings calls focused on how much they were going to have to spend to keep competing in this AI space. In another Reuters article, AI companies lose $190 billion in market cap after Alphabet and Microsoft report.

Now, even if the market is getting a little bit concerned about the high costs of competing in AI, that doesn't mean that they're not spending handsomely on executives who can help lead the battle. Fortune reports that businesses are scrambling to get chief AI officers, with the average compensation package for that position well over a million dollars. The New York Times wrote a similar piece, the hottest job in corporate America, the executive in charge of AI.

The subheader for that piece from NYT is, many feared that artificial intelligence would kill jobs, but hospitals, insurance companies, and others are creating roles to navigate and harness the disruptive technology. In 2022, Glassdoor saw just 19 people join a forum on their site that had the title of chief or vice president of AI. In 2023, that number jumped to 122 people.

Now, of course, another role in artificial intelligence is around AI safety. However, interestingly, Google has recently split up the team that's focused on responsible AI. Wired writes, a crucial team at Google that reviewed new AI products for compliance with its rules for responsible AI development faces an uncertain future after its leader departed this month.

The team in question was known as Resin, Google's responsible innovation team. It was located inside the Office of Compliance and Integrity, which was in the company's global affairs division. According to Wired, it reviewed internal projects for compatibility with Google's AI principles and conducted over 500 reviews last year. Now, the founder of that team and its leader was a woman named Jen Janai. And in the wake of her departure from Google, Google split her team of 30 people into two.

Wired writes some will remain in place while others were transferred to Trust and Safety, which fights abuse of Google services and also resides in the global affairs division. At this time, no one appears to have been laid off.

Now, of course, on the one hand, this could be nothing. This could just be internal shuffling. And just because something changes doesn't mean that it's necessarily for the worse. However, anything relating to AI safety or responsible AI development is so hot button right now that you can see people getting jumpy around the news, even if it doesn't necessarily turn into anything important.

Now, one other Google departure story, the information reports that one Google DeepMind veteran is departing to launch a new AI agent startup. Ioannis Antinoglou is a longtime AI researcher who has now left to work on one of the most hot-button areas in AI development, which is, of course, the agent space.

If the current generation of artificial intelligence are tools that humans use, the idea of agents is that instead of having to use these tools, humans could just give AI goals, and it could figure out how to go execute the tasks to achieve those goals. A huge amount of experimentation has been done in that space over the last year, with some of those products like Multion starting to really come to market now.

Whether AI agents turn into a big part of the AI space this year remains to be seen. However, it seems very likely that a ton of energy will be poured into that space, and I, for one, am excited to see what comes of it. For now, however, that's going to do it for the AI Breakdown Brief. Up next, the main AI breakdown.

Welcome back to the AI Breakdown. Today we kick off with a really interesting report from FlexOS about the biggest and most used AI tools. Now, some of these are not going to surprise you, specifically ChatGPT being at the very top of the heap, but others on the list might surprise you.

To kick us off, let's just read through the top 10 by usage. Number one, ChatGPT. Number two, Bing AI. Perhaps no surprises so far. Number three, though, Grammarly AI. Number four, Character AI. All the way down at number five is BARD from Google. At number six is Brainly, a homework help site. Number seven is Course Hero, a tutoring site. Number eight is Replit, a programming and coding tool.

Number nine is Turnitin, an AI detection tool. And number 10 is DeepAI, an image generator. In other words, in the top 10 in this list, there's no Midjourney, which comes in at number 12, no Perplexity, which is at number 14, no 11 Labs, which is at 15, no Adobe Firefly, which is at 18, no Runway, which is at 22, no Claude, which is down at 27. So let's dig a little bit deeper into this report and see what is actually going on here.

Now in terms of methodology, FlexOS here is reviewing web traffic and search rankings. Here are the way that they describe their key insights from their survey. Their first insight: the talk about AI is writing and images, but usage is in AI chat buddies and homework help. They point out that three of the top 10 applications here were in education, and that in the social and chatting category, Character AI is a huge winner which takes fourth place across all platforms.

Now, this character AI thing is something that has come up basically on every study like this that we've seen. It may be because it's disconnected from a more adult use case and that this is where the younger kid usage of AI is coming into play. But it always shocks people to hear how used character AI is, especially relative to some of the other tools, which take, frankly, much more top billing on shows like the AI breakdown.

Now, at the same time, the fact that education applications of AI are highly used is probably a little bit less surprising. Indeed, one of the things that we've seen is it seems like a lot of ChatGPT's usage is also focused on education, given that it went down last year when the summer session hit and kids were not in school anymore.

There are debates to be had around whether that is a good or bad thing or what it says about the state of education, as I have profiled on this show before. I think it's quite dismissive to assume that people using ChatGPT for education just means that they're copying and cheating on homework. But ultimately, it's not surprising that young people are figuring out how to use and leverage AI to make school easier and more effective for themselves. The second big insight from the survey is, quote, AI tool usage is way higher than people expect, beating Netflix, Pinterest, and Twitch.

Now, this is based specifically on how many monthly web visits ChatGPT has. So by proxy, the New York Times has around 609 million web visits per month. Twitch has a billion. Microsoft has a billion. Pinterest has 1.1 billion. Netflix has 1.5 billion. And ChatGPT has 1.7 billion.

Now, while this is very impressive, I think obviously these numbers would probably look a little bit different if you talked about time spent on platform. You have to think that Netflix, for example, and Twitch would both be way over ChatGPT in terms of actual time spent on site. But still, 1.7 billion visits per month is a serious number. FlexOS's number three takeaway? AI tool usage is surprisingly concentrated around a few winners. As they point out, ChatGPT is the leader by a mile, taking over 50% of the billions of monthly visits.

And when you take into account the other general GPTs like Bing, Bard, Clod, and Copilot, that represents 66% of all generative AI usage.

Now, they bring up a concern that this could be headed towards monopolistic power, and that certainly is a concern, although one of the interesting dynamics is that for the entirety of last year, GPT-4 has just been unquestionably state-of-the-art and better than everything else. And as the band of quality compresses, I feel like it's pretty unlikely that people have some major brand affiliation with ChatGPT that they're not willing to throw out if another tool suits their purposes.

Now in terms of other categories, writing and editing represented 9% of usage, education represented 6% of usage, social and characters represented 5% of usage, and everything else represented 13% of usage. One thing that is notable from this and shows just how tricky some of this type of methodology is, by the way, is think about something like mid-journey.

This survey is, of course, measuring web traffic, but Midjourney is used primarily, in fact, almost entirely, in Discord, which presumably isn't captured in this, although if it is, apologies to the FlexOS folks for not realizing that. But usage inside third-party apps like Discord could change these results at least in some ways.

Still, I think that broadly speaking, it probably is reflective that the general LLM usage is overall the biggest category, but I wouldn't be surprised to see if things like image generation actually represented a bigger slice of the pie than they're represented here. We also, for example, don't have a breakout of how much DALI is contributing inside the context of ChatGPT, given that those experiences are all managed from the same place.

Now, number four is really interesting. You remember Apple App Store's old campaign, there's an app for that. Well, the number four conclusion from FlexOS is there's an AI for everything. They note that more than 60 AI platforms already have over 1 million monthly visits. As a content creator in this space, this is certainly something that I noticed as well, that people right now really are exploring how to apply AI to every single thing that you can imagine.

One of the interesting dynamics and questions is to what extent ultimately there will be super apps like ChatGPT that dominate everything, or whether these more discreet and focused applications actually can figure out how to build sustainable models where the landscape is a little bit more specialized.

Some of the long tail categories that they recognize from the top 150 AI applications in terms of usage include generating video and animation from text, producing audio voice clones, taking meetings, producing podcasts, creating social content, organizing the workday, training employees, creating logos, answering customer questions and analyzing data. FlexOS's number five takeaway, a new generation is growing up with AI.

Basically, they're pointing out that among these top 50 tools, many of them like Brainly, Course Hero, and Character AI are disproportionately used by the youngest cohort. For example, 56.7% of Character AI's users are 18 to 24. Another 22.7% are 25 to 34. In other words, 80% of their users are less than 34. For FlexOS, they suggest that that means that employers should start to think in terms of an AI-literate employee base.

Now, one of the interesting things is when you take away general GPT or general LLM usage, how concentrated the usage of AI still is around things that you can use those general LLMs for. For example, when they take away general GPTs or general LLMs, among the remaining categories, writing and editing makes up 23% of usage, education makes up 16% of usage, and research makes up 12% of usage.

So right there you have over 40% of usage that's not general LLMs that actually also is probably done in general LLMs as well.

One side note, which isn't a takeaway that they had, but is something that I think is notable. While a lot of these categories are dominated by totally new players like ChatGPT, it's interesting how certain players have been able to transition to this new AI world really successfully. The fact that Grammarly, for example, has such dominance in the writing and editing category is testament to their ability to, one, see these trends early and start to build towards them, and two, just their ability to adapt to a changing environment.

Now, this education category, I think, is fascinating because on the one hand, it's already the number three category overall. But again, that's excluding just using ChatGPT for education, which I would wager to bet is a meaningful part of its usage already. I think that the type of applications that we're seeing inside education right now with Brainly and Course Hero are just the very scratching of the surface. And that this is one of the areas that is going to be most thoroughly disrupted by AI across a just huge array of dimensions.

If you're looking for a category that's one to keep an eye on for future updates, social and characters is definitely fascinating. Like I said, it's the one that's the most natively out of my experience thinking about this primarily from a professional standpoint, but it's clearly got a ton of usage and could represent something meaningful going on.

Anyway, overall, it's a really fascinating survey, even if you take into account that it's got some limits in its methodology. So great job to FlexOS for showing this, and I hope this was interesting for you. Until next time, peace.