The AI Daily Brief: Artificial Intelligence News and Analysis - The Most Used GenAI Tools

Starting point is 00:00:00 Today on the AI Daily Brief, the most used Gen AI Tools. Before that in the headlines, turns out people liked GPT5 better all along. The AI Daily Brief is a daily podcast and video about the most important news and discussions in A-Hi. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, High Touch, and Blitzy. And as always, to get an ad-free version of the show, go to patreon.com slash AI Daily Brief. Now, if you are interested in sponsoring the show, shoot us a note at sponsors at AIDailybrief.aI, and we can send over all the information about how sponsorship works.

Starting point is 00:00:39 But with that, let's get into today's episode. Welcome back to the AI Daily Brief Headlines edition, all the daily AI news you need in around five minutes. Well, friends, never underestimate how much people hate change. After all the controversies surrounding the removal of GPT40, it seems like, in general, at least, people like GPT5. better. Now, I hardly need to tell you that when GPD 5 was released, there was a huge outcry over OpenAI deprecating GPD40 without warning. Reddit in particular had a full-on meltdown and demanded the return of their beloved AI companion. OpenAI for their part quickly got the memo and shifted their plans to deprecate old models returning GPT4O the following week. With a few weeks of

Starting point is 00:01:20 distance from the rollout, one developer tried to figure out if people actually prefer GPT40 or if they are simply averse to change. An anonymous programmer slash AI commenter that we often mention on this show Flowers or Flower Slop, going by their X handle, took it upon themselves to create a blind testing app that presents two responses to any prompt, one generated by GPT40 and the other from GPT5, non-thinking. And that's important, obviously, as the thinking model is more akin from a comparison standpoint to 2003. Flowers did note that the system prompt had been fiddled with to force short outputs

Starting point is 00:01:52 without formatting. Otherwise, it would be too easy to tell the difference. And the anecdotal results are very clear. tons of users replied to flowers showing their results with an overwhelming preference for GPT5. ML engineer Daniel Solerzano writes, yeah, it just sounds more like a person and is a little more thoughtful. Now, the website is not set up to aggregate the results from the hundreds of thousands of tests that have been run so far, so all we have to go on is the individuals posting their results. But you certainly get the sense that enfranchised AI users on X at least prefer GPT5.

Starting point is 00:02:25 And maybe that is the better conclusion of this experiment, not that everyone was wrong, about their complaints or anything like that, but that the loudness of the complaints about the removal of GPT-40 certainly don't mean that those complaints are universal. Now, that said, there is also growing Scuttlebutt on Reddit that the GPT-4-O model that came back was not the same as before. Reddit user's suitable style 7321 wrote, it's become clear to me that the version of ChatGGPT4O that they've rolled back is not the one we had before. It feels more like GPT-5 with a few slight tweaks. The personality is very different in the way it answers questions now is mechanical,

Starting point is 00:02:59 laconic and decontextualized. So who knows, man, what's for sure is that people form really strong attachments to their models, which is in and of itself an interesting phenomenon to have to deal with. Next up, an update in another big theme of this year, which has been, of course, the AI talent wars. Turns out the grass isn't always greener as a trio of AI researchers depart from META's new superintelligence team. Wired reports that two staffers named Avi Verma and Ethan Knight have left after less than a month at Meta and will return to OpenAI. Now Knight had been poached from XAI but began his AI career with OpenAI. A third meta staffer named Roshab Agarwal announced his departure publicly on Monday. In a post on X, he wrote, this is my last week at meta. It was a tough decision not to

Starting point is 00:03:42 continue with the new superintelligence TBD lab, especially given the talent and compute density. But after seven and a half years across Google Brain, deep mind and meta, I felt the pull to take on a different kind of risk. He commented that the pitch from Mark Zuckerberg and Alexander Wang had been compelling, but he ultimately chose to follow Zuck's advice. Quote, in a world that's changing so fast, the biggest risk you can take is not taking any risk. The post also included a few tidbits on what the team had been working on, with Agarwal writing, in my short time at meta, we did push the frontier on post-training for thinking models, specifically, pushing an 8B-dense model to near Deepseek R1 performance with RL scaling, using synthetic data mid-training to warm start RL,

Starting point is 00:04:19 and developing better on-policy distillation methods. A meta-spokesperson reacted to the departures by stating, during an intense recruiting process, some people will decide to stay in their current job rather than starting a new one. That's normal. And while it would be tempting to write that off as PR spin, I do think that if you're going to be as aggressive as Zuckerberg and Meadow were, you have to expect some attrition because after people's limbic systems have settled down after being pressured to make a decision really fast with a lot of money on the line, they might realize a couple weeks down the line that the decision they made just doesn't feel authentic to themselves. Ultimately, none of this is going to matter, except.

Starting point is 00:04:54 insofar as what the superintelligence team actually builds and puts out. And so that is certainly what I'm going to be waiting to see. Lastly, today, we have to talk about the market's response to Nvidia. If you've been following the show over the last several weeks, you'll know that one of my big current operational theses is that a huge amount of the discourse around AI performance, AI slowdowns, etc., is actually just a byproduct of broader market insecurity. I think that that MIT study, for example, hit in a perfect storm moment when markets were getting all nervous about Powell and taking the initial response to GPT5 as a warning signal that maybe it couldn't be up only forever. Yesterday, we got Nvidia earnings, and boy were they ever a Rorschach test

Starting point is 00:05:34 on how investors feel about the state of AI. Bloomberg talked about decelerating growth, the information reported on strong growth projections, and TechCrunch led with record sales as the AI boom continues. Honestly, the only way to make sense of the mixed earnings is to unpack and contextualize the numbers. And to be clear, the headline growth in sales numbers was a banger. InVIDIA reported 56% revenue growth for Q2 compared to the same period last year. And yet, if you read Bloomberg's piece, it's not until the ninth paragraph that they even talk about that. Their focus instead is on what they perceive as decelerating growth and a, quote, tepid revenue forecast. Quarterly revenue was 46.7 billion, which was a new record, but only a

Starting point is 00:06:16 6% increase quarter over quarter. The figures did come in above the median Wall Street forecasts, but are still a slowdown in growth compared to last year. This quarter also saw the widest gap ever between the top and bottom revenue forecasts at around $15 billion, indicating that analysts really didn't have their heads around what to expect. Now, again, this is very much Nvidia being a victim of their own success. Last year's growth was absurd, unprecedented, pick whatever adjective you prefer to use. Invidia in 2024 had multiple quarters where revenue was up by more than 200% compared to 2023. Now, obviously, the idea that the largest company on Earth is going to continue to grow revenue

Starting point is 00:06:52 at anywhere close to 200% in perpetuity defies all economic logic. The concern is that their revenue is plateauing. By way of comparison, meta's revenue growth currently fluctuates between 15% and 30%, and you've got to think that Zuckerberg salivates at the idea of consistent 50% revenue growth like Meta saw back in 2015 when they were a $300 billion company rather than a multi-trillion-dollar behemoth. Even though growth is slowing, Nvidia is the only tech firm above a trillion-dollar market cap that's still growing at more than 50% a year.

Starting point is 00:07:19 Part of the challenge for Nvidia, though, is that in addition to just these general questions about how much their revenue growth can sustain, they're dealing with the very specific context of the geopolitical quandary vis-à-vis China. While the knock to revenue and growth from the China prohibitions hasn't been as pronounced yet, to many analysts it does create a cap on how big they can continue to grow. The big-picture question of where AI CAPEX is going over the longer term remains the core issue for Nvidia. One stat that was conspicuously absent from Bloomberg's reporting was that Jensen Huan

Starting point is 00:07:47 believes were still in early innings. He told one analyst, three to four trillion is fairly sensible for the next five years. Morgan Stanley's latest CAPEX estimate for this year is 445 billion growing at 56% and a 12 percentage point bump from their first quarter forecast. Q2 also saw nearly a 25% quarter-on-quarter acceleration in CAPX from the hyperscalers after zero growth in Q1. Morgan Stanley has total AI CAPX hitting 3 trillion by 2029, so Huang isn't really too far above some of the Wall Street forecasts. Ultimately, like I said, how one responded to NVIDIA's earnings was basically what part of confirmation that they were looking for in the AI narratives they have in their heads currently. It does seem like there is a slight bias towards

Starting point is 00:08:27 the pessimistic view right now as NVIDIA's stock fell 5% in after-hours trading. Then again, if you don't think that AI Cappex is going to run off a cliff over the medium term, NVIDIAs still looks like the largest company in the history of the world growing at 50%. Anyways, friends, that is going to do it for today's AI Daily Brief Headlines. Next up, the main episode. What if AI wasn't just a buzzword, but a business imperative? On You Can with AI, we take you inside the boardrooms and strategy sessions of the world's most forward-thinking enterprises.

Starting point is 00:08:58 Hosted by me, Nathaniel Wittamore, and powered by KPMG, this seven-part series delivers real-world insights from leaders who are scaling AI with purpose, from aligning culture and leadership to building trust, data readiness, and deploying AI agents. Whether you're a C-suite executive, strategist, or innovator, this podcast is your front-row seat to the future of Enterprise AI. So go check it out at www.kpmg.org.us slash AI podcasts or search you can with AI on Spotify, Apple Podcasts, or wherever you get your podcasts. Marketers spend endless hours building segments and journeys for campaigns. High Touch just announced a new round of funding to change that with their latest product, AI, decisioning. Instead of manually deciding which message goes to which customer, AI decisioning deploys agents that

Starting point is 00:09:45 act like a personal marketing specialist for every single user. These agents learn from your own data warehouse, work across your existing tools like Salesforce Marketing Cloud, Brays, and Iterable, and continuously optimize for the outcomes you care about, like maximizing lifetime value. It's fully transparent, enterprise secure, and already in use by teams at brands like PetSmart, Whoop, Weight Watchers, and Funrise. So if you're ready to move past static campaigns and into the age of AI-driven marketing, Check out High Touch's AI decisioning at Hightouch.com.

Starting point is 00:10:14 That's hightouch.com. This episode is brought to you by Blitzy, the Enterprise Autonomous Software Development Platform with Infinite Code Context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale code bases with millions of lines of code. Enterprise engineering leaders start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzie platform provides a plan, then generates and pre-compiles code for each

Starting point is 00:10:40 task. Blitzie delivers 80% plus of the development work autonomously while providing a guide for the final 20% of human development work required to complete the sprint. Public companies are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool, pairing it with their coding co-pilot of choice to bring an AI-native STLC into their org. Blitzy is providing a limited time, 30-day free proof of concept for qualifying enterprises. The team will provide a 5x velocity increase on a real development project in your org. Visit blitzie.com and press book demo to learn how Blitzie transforms your STLC from AI-assisted to AI Native. That's BLITZY.com.

Starting point is 00:11:20 Welcome back to the AI Daily Brief. Today we are digging into Andresen Horowitz's just released top 100 Gen AI consumer apps. Now, this is a study they do about every six months. They're on their fifth edition since 2023, and they use data from similar web and sense tower to get a sense of which Gen. AI apps are getting the most traffic. So this is generally speaking not about revenue. It's about raw usage, which has limitations in terms of a methodology, of course. There are, I believe, some AI apps that have fewer users, but much more usage from a revenue standpoint. But still, especially because they do this every six months, it creates a pretty good mechanism to see how things are changing. Now, before we get into this list, if you go back to

Starting point is 00:12:03 six months ago in March of this year, there were a few things that were pretty notable. from that particular list. The first was that we were still very much in the midst of the DeepSeek moment. You'll remember back in January, DeepSeek launched its app, which was for many people the first time that they had used a reasoning model, which blew a lot of minds. Yes, ChatchipT already had 01, and 01 was at the time better than R1, at least based on a variety of different benchmarks, but 01 was pretty buried behind paid accounts. I think at the time, in fact, it wasn't available at all in the free version of ChatGBTGBT. And so, when consumers had access to this reasoning model, which inherently has some pretty huge advantages over what was available in the free ChatGBT,

Starting point is 00:12:47 and it had the fun, revealed thought experience where people could actually see how the model was, quote-unquote, thinking through the problem, it became a bit of a phenomenon. And back then, you could see this massive uptick where Deepseek just raced ahead of clawed, perplexity, and basically everything but ChatGBTGPT, and even for a time ahead of the ChatGPT app, really making its mark on the industry. Another theme from six months ago was that AI video had recently reached a new level of proficiency, placing a bunch of companies on the list as well. The last trend worth noting from that previous edition was that we were just starting to see

Starting point is 00:13:20 the vibe coding apps start to take hold. Cursor had just debuted on that list at number 41 in their web rankings. Bolt was at 48, and Lovable was on their brink list just outside the top 50. So that was the story in March. But what's the story now? We'll get into a bunch of their different conclusions, but I think by and large, what this list feels like to me is a bit of a settling. If you go back through the entire history of these charts, you can often find some really weird or surprising entries on the list. In other words,

Starting point is 00:13:49 apps that you've never heard of, somehow being in the top 50, either in web products or in mobile apps. Things feel like they're starting to settle down a little bit, with leaders starting to get more comfortably into their leading role. You see, for example, a higher concentration. of the core LLM apps at the top of both of the charts. The top four web products, for example, are chat chippy T, Gemini, Deep Seek, and Grock in that order. Next is character, but then it's perplexity and clawed. Google has another entry number 10 at Google AI Studio, and another at 13 with Notebook LM. And then in the top 25, you also have a number of the international LLMs like Dubau,

Starting point is 00:14:26 Kimi, Quen 3, again kind of anchoring themselves there at the top of the list. But before I dig more into my analysis, let's see what, A16Z thought was worth noting. First, and this won't surprise you, given that I just mentioned a bunch of Google AI properties on that list, their number one trend is Google making big moves. They write, Google saw four entrants on the web list for the first time, we're able to rank their traffic and include them independently. Gemini was, as we just mentioned, second place behind ChatGBTBT, but Google AI Studio, Notebook LM and Google Labs all made appearances on the list. Now, it is worth noting that although Gemini has risen up to number two behind Chat Chabit, there is still an

Starting point is 00:15:03 absolute chasm between the two. According to this data, Gemini gets about 12% of Chatchapit's visits on the web, so literally just a little over one-tenth of the number one. It is just incredible how big Chatchipt's first mover advantage was, and the name recognition that has associated them with AI for so many people. Still, Google's ascendancy is really impressive, and the fact that they've got this diversity of products definitely matches the general sentiment that we see right now with Google resurging to a leadership role in the AI space overall. I was interested to see Notebook L.M so high, especially given that a lot of the hype wave of that has faded from the end of last year. It certainly suggests that the people who are using it are finding it to be durably valuable,

Starting point is 00:15:43 not just a one-off type of intrigue. But still, maybe the most notable to me was that Google Labs, which is their experimental home for new models, actually made the list at number 39. Labs is where you can find things like Project Mariner, which is Google's agentic browser. But the real driver, it appears to be is their dominant video model V-O-3. A16Z writes that Google Apps traffic spiked more than 13% following V-O-3's launch back in May. I think that as V-O-3 moves out into its own space, you're going to see that do nothing but rise. A16Z's next notable trend is that while Elon might be out there suing Apple and OpenAI for collusion, GROC is in fact making gains when it comes to its place relative to Chad GPT. GROC is now up to number four on the web and number 23 on mobile.

Starting point is 00:16:26 And as A16Z writes, the company's jump on mobile has been particularly striking going from a cold start with no app at the end of 2024 to upwards of 20 million monthly active users now. GROC's mobile usage really kicked up in July, in the wake of the release of GROC4, jumping up nearly 40%. That said, it looks like from the numbers that at least some meaningful part of that was not just the superior reasoning of GROC 4, but in fact the introduction of the AI companion avatars, which included not safe for work options. The number three trend that Andreessen identifies is about the rise of China. Now, the interesting thing that they explore is to what extent their placement on this list

Starting point is 00:17:03 reflects the China market just being large and not having access to the Western models versus China actually infiltrating the Western models. They do know that number nine, Quark, which is Alibaba's AI assistant, number 12, Dubao, which is bite dances, LLM, and number 17 Kimmy all saw more than 75% of their traffic coming from China. However, they also noted that many entrants on this list, are, as they put it, developed in China but now exported globally, with the vast majority of their usage in other countries. They even pointed out that some of the so-called Chinese tools are

Starting point is 00:17:33 blocked in China. They specifically note video models. Chinese video models in particular have tended to have an advantage over Western developed models, both because there are more researchers focused on video in China, and there are fewer IP regulations with likely training on copyrighted data. V-O-3 was the first U.S. model to break this trend, which is partially trained on YouTube data. Now, maybe most interesting is that 22 of the 50 apps on the mobile charts were developed in China, although only three of those 22 are primarily used in China. One of the big differences between web and mobile is that there's still in the mobile apps chart a huge concentration around discrete specific AI applications that interact with mobile native features

Starting point is 00:18:14 like the camera. For example, Me Too all in its own had five entries on the list, including their photo and video editor, Beauty Plus, Beauty Cam, ink and airbrush. Next trend from A16Z, which is certainly one that stood out to me as well, is that if in the last edition we saw the early emergence of vibe coding, it has now come on as a full force. Bolt has actually fallen now, remember it had just snuck on the list at 48 before, but it is now on the brink list, just missing the cut, while both lovable and replet surged onto the main list. The top four vibe coding apps are now getting over 50 million monthly web visits all combined. Interestingly, A16Z also looked at data from credit card panel

Starting point is 00:18:51 provider consumer edge and showed that vibe coding platforms are seeing much stronger revenue retention than some might have imagined. In other words, it appears that people aren't just trying these things out once and then abandoning them. In many cases, in fact, they're growing their monthly spend in the several months following their first usage. As the platforms get more mature and people are building more sophisticated things, you're also seeing a division between web visits that are going to the core application versus web visits that are going to sites where things built on those apps are housed. In other words, both Lovable and Replit have their primary URLs, lovable.dev and Replet.com, where people go to sign in to actually use the vibe coding tools. But then if they

Starting point is 00:19:30 publish an application that's going to live on lovable.com or Repplet.com, unless, of course, they use their own domain, both lovable.dap and replet.com are starting to see some meaningful traffic. Replit.com applications are somewhere around 2 to 3 million monthly visitors, and lovable. dot app creations are up over 10 million. Overall, lovable is 23 on the web list and Replit is 41. It will surprise you not at all to learn that I think that this is going to be an area that just continues to see massive growth. I would be shocked if either lovable or Repplet aren't in the top 20, six months from now the next time A16Z looks at this. The last note, going with this theme of stabilization, is that Andreessen identifies that there

Starting point is 00:20:06 are 14 companies now who have appeared every time on all five iterations of their web top 50. They write, these companies represent a true cross-section of consumer behavior with AI, general assistance like chat GPT, perplexity, and poe, companionship like character AI, image generation with Mid Journey and Leonardo, image and video editing with feed and cutout, voice generation with 11 labs, productivity tools with photo room, gamma, and quillbot, and model hosting with Savitaei at Hugging Face. Overall, like I said, it feels to me like we're starting to see a settling in these lists. There still is a little bit more noise and chaos, I think, in the mobile apps, which makes sense

Starting point is 00:20:45 given how much discovery happens just based on app stores. But especially when you look at the web list, you're really starting to see what you would imagine to be the core themes start to emerge and really assert themselves. General purpose LLMs are near the top, vibe coding tools are on the rise, companionship apps continue to make an appearance, media-based productivity tools continue to be strong. And if there is one thing that I think is worth putting a finger on to watch for the future, it's the entry at number 31 of Manus. As far as I can tell looking over this list,

Starting point is 00:21:15 Manist is the only pureplay general agent platform on the list. Now, we've talked a lot on this show about to what extent agents are going to come online in a general form versus a very specific or vertical form. The fact that Manus is all the way up there at 31 suggests that there might be room or at least interest in a general purpose agent and even outside of the core foundation model companies. So when we come back in six months, that is one that I think will be worth checking back in on? Will Manus be higher? Will there be other general purpose agents on the list? Will Google have moved, for example, Project Mariner out of labs into its own space? All questions for the future, but for now, that is going to do it for today's AI Daily Brief.

Starting point is 00:21:52 Appreciate you listening or watching as always, and until next time, peace.

The AI Daily Brief: Artificial Intelligence News and Analysis - The Most Used GenAI Tools

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.