ChatGPT Search and the AI Search Arms Race

Today on the AI Daily Brief SearchGPT is now live and the AI search wars are heating up. Before that in the headlines, Amazon is getting agentic and other news from hyperscaler earning season. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show notes. We are fully in earnings season right now and have a bunch of stories from that as part of our headlines Today. Kicking off Amazon is looking to give agency to Alexa during yesterday's earnings call, Amazon CEO Andy Jassy hinted at an agentic version of the company's virtual assistant. He said, I think the next generation of these assistants in generative AI applications will be better at not just answering questions and summarizing index and aggregating data, but also taking actions. And you can imagine us being pretty good at that with Alexa. Jassy confirmed that efforts are still ongoing to quote re architect the brain of Alexa, a process that includes a new set of foundation models that the company is planning to reveal in the near future. Efforts to build an AI powered Alexa were first revealed back in 2023 with prototypes using models supplied by Anthropic. In practice, we haven't seen much of this iteration of the project, save for an unreleased version that struggled to turn on Smart Lights and was painfully slow. The new build, which is internally codenamed Remarkable Alexa, which is so painfully cringe, will reportedly cost between $510 per month alongside a stripped back free version. It was originally slated for an October release, but Bloomberg reports that the timeline has shifted into next year. Earlier this week, Bloomberg chronicled some of the difficulties that have held back the launch of an AI powered Alexa. They noted that Amazon is facing a fundamentally different problem than other AI labs. Bloomberg writes users toying with ChatGPT expected to make mistakes. If Amazon turned on its LLM brains and Alexa began spewing provocative answers, it it could turn into a fiasco for Jassy given the huge portion of kids and families that use Echo hardware. And this is a really interesting counterpoint to the fact that these big tech companies have such big distribution. Yes, Alexa has a huge number of built in devices, but the stakes are also higher. The expectation and how accurate AI has to be is just higher when it comes to a device that's already installed and integrated into your life than from some new thing that you're just trying to figure out how much you're going to trust it or not. Overall, big tech earnings this week revealed a lot about how Wall street is feeling about AI. All of the public hyperscalers have now reported quarterly financials, and the response was somewhat of a mixed bag. Google saw a positive response earlier in the week, with investors liking strong growth in cloud revenue and Sundar Pichai statement that one quarter of internal code is now being generated by artificial intelligence. On Wednesday, though, Meta and Microsoft each saw negative responses for their own reasons. For Meta, the story continues to be about massive investment with too little to show for it. The company is set to hit $100 billion in costs across all divisions this year, with around 6 billion going to AI research and infrastructure. The company is generally showing solid growth with quarterly revenue at a record high $40 billion, up 19% from this time last year, slightly slower than the 22% growth they posted last quarter, but still a solid pace. The escalating investment cost is what's starting to concern investors. Remember, Meta is just coming out of a period of intense belt tightening. The company was criticized for overspending on their virtual reality division, which was such a large bet that it of course prompted a name change. That division has struggled to deliver a return on investment and is still posting over $4 billion in quarterly losses. Zuckerberg was forced to face that music in late 2022, signaling a year of efficiency, his words, conducting multiple rounds of layoffs and cutting way back on investment spending. That era is clearly over, with Zuckerberg stating during the earnings call, our AI investments continue to require serious infrastructure and I expect to continue investing significantly there. The company guided quote a significant acceleration in infrastructure expense growth next year as we recognize higher growth and depreciation and operating expenses of our expanded infrastructure fleet. Meta is somewhat unique among the hyperscalers in that they don't have a cloud division that is clearly reaping the benefit of AI adoption. Instead, the company needs to show a return on this massive spend through metrics like time on platform and advertising efficiency. They did attribute an 8% increase in engagement time on Facebook and 6% on Instagram to AI powered feeds, but investors are still skeptical. Brian Mulberry of Zack's Investment Management said there's been an enormous amount of capital spend on developing AI, looking very specifically to see who is monetizing it, who's actually giving a return on investment, and what does it look like. Microsoft, meanwhile, is in a slightly different position. The company showed strong growth in their cloud division, which is capturing the Lion's share of AI gains. The group grew by 33%, far above expectations. Capex for the quarter were almost 15 billion, which came in above expectations and was a 50% jump from last year, CEO Satya Nadella said. Our AI business is on track to surpass an annual revenue run rate of $10 billion next quarter, which will make it the fastest business in our history to reach this milestone. Still, the company also admitted that Azure Cloud is basically at capacity and the company needs to construct more data centers to expand. Apple was a little lackluster with the stock falling in after hours trading. They reported revenue at an all time high propelled by a modest rebound in iPhone sales growing by 6%. But the problem is that Apple doesn't see a big jump in sales coming. They guided low to mid single digit sales growth for the December quarter and these were supposed to be the two quarters where Apple demonstrates that AI can drive an upgrade cycle. So far consumers have not responded to Apple Intelligence as a must have feature, but as we've talked about a number of times, features are barely there so far. One thing to note is that with all of these companies, the results were extremely positive by traditional metrics. Basically, the bigger shift is what Wall street expects and how much it's about real ROI now versus the promises of future returns. Still, as much as Wall street is in small ways punishing companies for high capital expenditures, it's pretty clear whether overinvestment or underinvestment is worse. Shana Sissel, the president and CEO of Ban Rion Capital Management, said the CEO of a tech company is much more likely to get fired for not spending enough on AI versus spending too much. Because right now that's the deciding factor. Anyways, that is going to do it for today's AI Daily Brief Headlines edition. Next up, the main episode. Today's Episode is brought to you by Venice. Venice is a private, uncensored, generative AI app. It accesses open source models to enable text, image and code generation without the fear of being spied on or having your data exploited. Discuss anything with Venice without concern about it being monitored, sold or given to advertisers and governments. Venice is different because your conversations and creations are kept securely within the browser, never stored or accessible by Venice. Unlike other AI apps, Venice won't tell you what's okay to say or not. Venice won't patronize you. It simply provides direct access to machine intelligence. No topics or off limits, no ideas are taboo. With Venice, you're in control of the AI as you should be. Pro subscriptions are available for $49 a year or $8 per month. AI Daily Brief listeners receive a 20% discount on Venice Pro. Visit Venice AI NLW and enter the discount code NLW Daily Brief. That's NLW Daily Brief. All one word. Today's episode is brought to you by superintelligent. Every single business workflow and function is being remade and reimagined with artificial intelligence. There is a huge challenge, however, of going from the potential of AI to actually capturing that value. And that gap is what superintelligent is dedicated to filling. Superintelligent accelerates AI adoption and engagement to help teams actually use AI to increase productivity and drive business value. An interactive AI use case registry gives your company full visibility into how people are using artificial intelligence right now. Pair that with capabilities building content in the form of tutorials, learning paths and a use case library, and superintelligent helps people inside your company show how they're getting value out of AI while providing resources for people to put that inspiration into action. The next three teams that sign up with 100 or more seats are going to get free embedded consulting. That's a process by which our super intelligent team sits with your organization, figures out the specific use cases that matter most to you, and helps actually ensure support for adoption of those use cases to drive real value. Go to BeSuper AI to learn more about this AI enablement network. And now back to the show. Today we are discussing OpenAI's ChatGPT search, but really what we're discussing is the absolute search wars that have now engulfed the Internet. It has been a very long time since anyone actually factually practically threatened Google's dominance of search. And yet now it is so clear that whether it's Google that wins or not, what search means is going to change in a fundamental way. We've got perplexity out here innovating, really pushing the boundaries of what the summary model of search can be. We've got Google trying to catch up with their overviews and at the same time trying to figure out how they can do that without disrupting their own business model. And now we have OpenAI entering the ring as well. In an announcement post, OpenAI said that their new ChatGPT search service is designed to provide fast, timely answers you would have previously needed to go to a search engine for. In other words, the idea here is that rather than having to cruise around and click on links and try to find the right reference point, you instead just get a summary that answers the question right there for you. OpenAI's model is powered by a fine tuned version of the GPT4O model and can serve up photos and information from the Internet OpenAI gave examples of sports scores, news and stock quotes. And just like with perplexity, users can also ask follow up questions to refine their search. ChatGPT search produces both inline and sidebar attribution to news publishers and other data sources. OpenAI has licensing deals with and this is an important point. One of the things that OpenAI keeps saying as they sign these deals is that it's not about training data, it's about how they surface the most reputable sources in their search. For example, results related queries about the election will be directed to sources like the AP and Reuters. The service is now available to ChatGPT and team members on mobile and the web, with other subscribers gaining access over the coming weeks, followed by a rollout to free users. The company has also released a browser extension to allow users to make ChatGPT search their default search engine. OpenAI said they plan to keep improving search, particularly in the domains of shopping and travel. They also plan to add voice mode and access the O1 Reasoning models for what they call deeper research, speaking to the idea that this is where search is headed, a quote from Louis Dreyfus, the CEO and publisher of Lamond we are convinced that AI search will be in the near future and for the next generations, a primary way to access information. Partnering with OpenAI Positions Le Monde at the forefront of the shift. It allows us to test innovations at an early stage while safeguarding journalism's core values and integrity. Matias Sanchez, the SVP of Global Strategic Partnerships at Axel Springer, says, as AI reshapes the media landscape, Axel Springer's partnership with OpenAI opens up tremendous opportunities for innovative advancements. Together, we're driving new business models that ensure journalism remains both trustworthy and profitable. I do not think it's at all an accident that OpenAI is going to pains to highlight all of their partnerships with companies like the Associated Press, Conde Nast, Meredith, Financial Times, etc. Especially given that Perplexity is now starting to be awash in lawsuits and trying to catch up on their own licensing deals. OpenAI, for their part, are very clearly excited about this. Sam Altman tweeted, hey, I'm really sorry to keep hyping our own product, but you really should get ChatGPT and install the Chrome extension for search. I am cheerfully the first to admit when we ship something that isn't very good, but this time it's really good. On the positive side, some people noticed that it was really fast. AI entrepreneur Saliomar said. My only question about SearchGPT how on earth is it so fast? Seriously, what and how did they cook? It's nearly instant. OpenAI's head of applied research Boris Power, responded. Early feedback was that it's slow, so we put effort into making it super fast. To which Sully said, teach me the way Rowan Chung from the rundown writes first impression trying it feels fast. It even pulled our logo second impressions. He continues, Around 2x faster than perplexity Pro, but not as in depth. Rowan continues, I like the source tagging better than Perplexity, says the company name rather than a number. No related questions at the bottom like Perplexity, which is one of my favorite Perplexity features. Rowan then gives his power rankings My AI search rankings as of today. Perplexity Pro over ChatGPT search ChatGPT search over Google Search Google search over Bing. This doesn't mean ChatGPT search, Google or Meta can't eventually beat Perplexity, but it does mean the AI web search. Frenchie just got serious. A word of caution, however, he finally adds, ChatGPT search is hallucinating when you ask it for tweets. X is notorious for being hard for search engines, but Perplexity at least says it can't see content instead of making something up. And that indeed brings us to some of the negative comments. Hallucinations are very high among them. Professor Ethan Malik writes, finding very high hallucination rates right now. I'm sure it will improve. Investor Stephen Sinofsky says, made up stuff non reproducible answers slow. What more could you not want from search? Copenhagen Business School Professor Christian Hendrickson writes, I have searched GPT and my very first search thread was a failure. Later in the conversation it started completely hallucinating the newest dev diary. Perplexity is still my go to Benjamin Decracker, who it should be noted is @OpenAI competitor Xai now writes, I don't really get ChatGPT search. Tried multiple quote unquote searches and the results are very similar to just normal GPT results. Very text heavy. Was expecting a more search first experience. Now it's clear as you heard from all of these things is that the perceived battle is with Perplexity. Perplexity is in an absolute arms race to advance new features. CEO Aravan Srinivas tweeted yesterday, until now we mainly prioritized informational queries, but search is about anything you want to do. A navigational query is essentially a link in a sitemap. As an answer, we made it even easier to navigate the web. Especially useful when you set Perplexity as your default search developer Nick Dobos responds, this is huge. The only reason I didn't have Perplexity set as my default was the 25% of time I needed to click the top result and navigate to it. Testing catalog news, writes Perplexity, just got even faster. But what's super important, it is accurate. Perplexity1 Search GPT0 Overall, the sentiment is pretty positive. Alim writes. OpenAI's design team is really stepping up and crushing it with their OS experience. GPT search looks incredibly slick too, and with their strong brand position, they're well placed to capture a massive slice of the market, probably the most. And indeed, we're kind of dealing with layers of distribution here. Perplexity recently boasted of serving 10 million monthly active users, which is amazing for how fast they're growing. At the same time, that number is closer to a couple hundred million for ChatGPT, giving them undeniably an edge on the distribution front. Then of course, there's Google, who's serving up something like 8 billion queries a day, which means you have to think that if they get it together, they have such an inherent advantage going in. Regardless, it is going to be a super interesting battle, one that it's hard to imagine we don't benefit from as consumers. One more little piece of OpenAI news before we get out of here today. An interesting substory from a Reddit AMA with OpenAI CEO Sam Altman yesterday. Altman says that the lack of compute is holding the company back. He wrote, all of these models have gotten quite complex. We also face a lot of limitations and hard decisions about how we allocate our compute towards many great ideas. Of course, we've seen a wave of headlines over recent months about how OpenAI is trying to address this issue. On the training side, the company is partnering with Oracle to build one of the largest training clusters in the world, in Texas for Inference. They recently announced they would be using AMD GPUs and have plans to start designing and manufacturing their own chips, partnering with Broadcom and TSMC now. As an example of a feature being held up by a lack of compute, Altman referred to vision capabilities for ChatGPT's advanced voice mode, stating that it wouldn't arrive anytime soon. This feature was showcased during the demo in April, but didn't arrive when the rest of voice mode was rolled out in late September. Other big news from the AFA included other indefinite timelines on new products, an update to OpenAI's generative image model. Dall E has no timeline with Altman stating we don't have a release plan yet. Sora, the company's generative video model, has also been held back, with Kevin Wheel, the company's chief product officer, saying the delay is due to the quote, need to perfect the model, get safety impersonation, other things right, and scale compute. Although the AMA was filled with news about delays, Altman insisted we have some very good releases coming later this year. Nothing that we are going to call GPT5 though. So there you have it. That is the story from OpenAI and that will be the end of our daily brief today. Appreciate you listening as always and until next time. Peace. Sam.