EP 394: AI News That Matters - November 4th, 2024

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

ChatGPT's new search tool is out, but that was actually kind of scooped by Google. Meta's llama model is reportedly being used by the Chinese military. Microsoft and Google are expanding, experiencing AI delays. Anthropic, isn't they? Keep shipping and chat GPT won't be shipping GPT anytime soon. Yeah, a lot going on in AI news this week as always and we're going to get you caught up right now. What's going on y'? All? My name is Jordan Wilson and I'm the host of Everyday AI. Welcome. Thank you for tuning in. If you're listening on the podcast, appreciate you as always. Make sure to check out your show notes for for more on all of today's stories that we're going to be recapping. Live stream audience thank you. It's great to see you. I'm actually in a different country in Australia right now on an anniversary with my wife. So, you know, who knew shout out to Sydney apparently I was checking our podcast stats. Sydney is the number one city all time. So thanks for kind of keeping with today's audio and video issues. Not going to be the best. Trying very hard but coming to you live ish in the hotel room here. So sorry if there's a little bit of an echo. But enough of that, let's just get into what actually matters and going over the AI news and like I said, if you haven't already, please go to your everydayai.com Sign up for the free daily newsletter. We will be recapping all of these stories. All right, let's get straight into it. That first story ChatGPT has introduced its enhanced web search capability for users. So chatgpt has just rol out an upgraded search feature that allows users to receive timely answers with links to relevant web sources, significantly improver improving the user experience. So the new search functionality is available to ChatGPT plus and Teams users with plans to extend access to free tiers in the coming months. So the feature enables users to ask questions conversationally and receive responses that may include information sourced from the web, making it easier to obtain accurate and relevant answers. And then follow up questions will be considered in the context of the entire chat, allowing for deeper engagement and more tailored responses. So this is essentially what we heard of a while back when we heard of this Search GPT. So when OpenAI released that they did say Search GPT was a prototype and what was just released does not look exactly like this, but it does look like a much improved version of OpenAI's older browse with Bing search. So the new ChatGPT search, I've used it. I'm sure many of you have as well. If you haven't used it, it really just kind of looks and feels a little more like Google. Right. So that's what I think OpenAI is really making a huge play for. Because here's what people don't understand right now. You do not even need an account to use ChatGPT. You don't need to be logged in. You can literally just go to chatgpt.com and start using it. So you can't use this new Chat GBT search that way. Again, right now you have to have an account and it needs to be a paid account. But I think this is where OpenAI is actually headed right now. I don't think they're trying to compete with Perplexity necessarily. I think they're actually trying to compete with Google. I think it's a smart play because the new ChatGPT search is really, really good. So feedback from global publishers so far has been instrumental in shaping this feature, emphasizing the importance of maintaining journalism's core values and integrity. So that's another big piece here. OpenAI has a lot of deals with publishers that is allowing them to source this material more quickly. So yeah, it is blazing fast. This is faster than Perplexity and it is even faster than Google's AI search. It's very impressive to see what OpenAI has done so far with ChatGPT search and I'm sure we'll be talking about this a lot more. All right, well, they actually kind of got scooped by this though, because Google, literally right before this, they launched their real time search for Gemini AI, setting the stage for some intense competition in the AI powered search. So Google has introduced real time search capabilities, but for its Gemini AI platform, not for its Gemini chatbots. So the new feature is called Grounding with Google Search and it enables developers to enhance their AI applications with current information sourced directly from Google Search. So like I said, this is not the same thing as ChatGPT search or perplexity. This is for developers who are already maybe using Gemini, but to get real time access to Google Search, which I think is huge. Right. But again, if you're just using AI chatbots as a front end user or if your company maybe uses the Gemini API, but you don't need real time Search this may not impact you however, if you are using the Gemini API or maybe you want to learn it and you also need real time access to Google. This is huge. So you know, you have to give your kind of tip your hat to to Google. I've been saying this for a very long time. I'm not impressed at all with their Gemini Chat, but their back end tools for developers. Fantastic. So this new grounding with Google Search, pretty cool. So the service right now is priced at $35 per 1000 queries. So kind of pricey, right? But reflecting the high computing costs associated with this real time search functionality on top of the Gemini API. So Google' dynamic retrieval system scores queries based on their relevance to current events with scores ranging from 0 to 1 to optimize cost and response time. So yeah, there's a little bit of handling on the front end there. So this integration comes at a critical moment as Google reported $49.4 billion in search search advertising revenue in quarter three when they just reported their quarterly earnings, while also fac increasing competition from AI driven initiatives. And like we said, OpenAI's recent launch of ChatGPT search offers a consumer focused approach, whereas this new grounding with Google Search is really offering a similar capability, but just for developers. So both companies have secured licensing agreements with major news organizations, although the details of these deals remained undisclosed. And it's not with all the publishers. Right. Here's the big elephant in the room that mostly people aren't talking about, right? All these big large language model makers essentially just scrape all website content, not just those that they have licensing deals with. So it does seem like some in some unofficial testing early on that they seem to prefer or to serve answers from those publishers that actually have, you know, the results. So you know, a way you can test this, it's very unofficial. You can do a normal like Google Search as an example and see, you know, which of these partnership players that maybe have a result on, you know, the first page of Google if you're searching for something and then you can do that same search, you know, if you are using the grounding with Google Search or similarly, you know, do something like that with Chat GBT search and you will see that OpenAI and Google are seemingly, you know, favoring these companies that they have licensing deals with. But that doesn't cover everything, right? So don't worry, you know, if you're a publisher out there, you know, your results are still going to be coming up on ChatGPT search and this new grounding With Google Search. Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for chat GPT training for thousands or just need help building your front end AI strategy, or you can partner with us too, just like some of the biggest companies in the world do. Go to your everydayai.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen AI. All right, our next piece of AI news. Speaking of the big large language model makers, Claude just shipped a lot. All right, so Claude launched their dedicated Mac app. So they've introduced a dedicated Mac app for Claude, enhancing accessibility for users on both Mac and Windows platforms. So the Claude app is currently in beta and available for download to all users, whether they are on free or paid plans. So you gotta love that from Anthropic. So the app right now provides easy, easy access to Claude's latest model, which is Claude 3.5 sonnet new. Yeah, I know you can just. I call it 3.6, a lot of people call it 3.6. But this also is not the only update because there's also dictation within the app and also two other things from Anthropic. So they also released the new PDF processing system that employs a three phase methodology, which is text extraction, visual processing and integrated analysis, allowing for a more comprehensive understanding of documents. So what that means. Yeah, let me just recap that. So we have now the new number one, we have the new desktop app. Number two, there is dictation now so you can speak to Claude and you know, instead of having to type to it. And number three, it has this new kind of three pronged PDF support. And this is huge because by default most large language models do not handle kind of, I'll say, multimodal PDFs right. Even though they're not multimodal technically. Right. But essentially most large Language models don't do well if the PDF that you upload and you're asking it a question or you're asking it to read something back or summarize something, most do not do a good job with images, graphics, etc. So this is really what this update from Anthropic is all about. The the PDF update is this new kind of three phase methodology, text extraction, which is not new, but the big thing here is the visual processing and the integrated analysis. So I did do a quick kind of little tip trick and tutorial on how to actually do this in Chat gbt because by default Chat GBT doesn't do this very well. You do have that, you know, know your way around some basic prompt engineering. So love to see this update from Anthropic. And hey, livestream audience, let me know if you want to see maybe a tutorial on these new PDF features from Anthropic. We got so much news. So while Anthropic was announcing updates and just shipping, Google was announcing delays. All right, so Google has delayed its Project Astra AI launch until at least 2025. So Google's ambitious Project Astra, which was aimed at developing AI applications for real time understanding, will not be launched until at least 2025, as revealed by Google CEO Sundar Pichai during the company's quarter three earnings call. So Project Astra was showcased in a demo at Google's IE developer conference in May of this year, highlighting its potential to enhance smartphone capabilities such as recognizing surroundings and answering questions. So essentially what this Project Astra was, It was what ChatGPT's live kind of voice assistant will be or could be once it gets video capabilities. Right. So what they showed in the demo was, you know, someone kind of walking around Google's office asking questions about things on the desk and they can, you know, kind of use computer vision and see in real time. That's pretty interesting here that Google had yet another delay. You will also remember about a year ago, Google informed, infamously kind of showcased some of these features right when it first launched its Gemini kind of chatbot. But then they kind of later said, ah, yeah, we, I guess everyone else kind of said like hey, this isn't really true, this is not how it works. So Google did have a live working demo of Project Astra, which again, it can see in real time with video and then it is a voice assistant that you can talk to. So it did appear to be a working demo in May at the IO conference. Yet we might not even get it in 2025. Yes, it says 2025 at the earliest, so it might not even be until 2026. So this is another page out of the Google Playbook with, you know, a shiny demo, you know, lots of marketing, and then, you know, you might get it in three to six months or in this case even longer. So pretty, pretty shocking that we might not even get it in 2025. So initial reports suggested that a consumer focused agent experience was set to launch in December 2024 and it was capable of handling tasks like purchasing products and booking flights. However, we don't know if this report still holds true. We talked about this last week, week on our AI News that Matter segment. So we weren't sure if that is more similar to Claude's computer use. Right. So more of a kind of agent that you run on your computer or if this was supposed to be Project Astra, but now, at least according to Google's CEO, so from the source himself, we are not going to be getting Project Astra. So pretty interesting there a lot of delays, and they're not the only one because Microsoft also just delayed its. I don't even know if we can call it a popular feature because it has not been released yet. But Microsoft has once again delayed its recall feature for Copilot Plus PCs until December. All right, so at least we have a timeline on this one. So Microsoft has announced another delay in the rollout of its recall feature for Copilot Plus PCs, which is now set to be previewed with Windows Insiders by December. So the recall feature, essentially it is like history for your browser, but for everything, right? So anything that you do within Windows programs, it can kind of remember it and then you can talk to it like you would a large language model. And there's. When it was first announced, I think there was both a lot of optimism, but also a lot of skeptic like, like skepticism, right? People are like, oh, this is a privacy nightmare. And I think Microsoft did a good job in addressing those concerns, but again, we are seeing more delays. So the recall feature was originally scheduled to be tested with Windows Insiders in October, but has, like we said, just been postponed till at least December for further refinement. So like I said, recall it captures screenshots of user activity on Copilot plus PCs and it aims to enhance product productivity by allowing users to search and retrieve their past actions. So the features development has been complicated by security issues leading to an overhaul that included making recall an opt in experience and ensuring a complete uninstall option for users. So yes, whether this was a kind of feature that was added or a security hole that Microsoft did plug but you know, now it is an opt in feature, whereas before it was presumed to be kind of a feature that that was on for everyone and then you had to turn it off if you did not want it. So Microsoft did clarify this a couple of months ago and say it is an opt in experience and also it does give the option for a complete uninstall for users. There's also been a little more confusion with some mistakenly believing Recall was being installed on all PCs running Windows 11. But that is not the case. It is is just for those Copilot Plus PCs. So these are for the newer Copilot Plus PCs that have the, you know, very powerful CPUs, GPUs and also the NPUs which allows you to run kind of this edge AI or on device AI. All right. So yeah, Microsoft made a ton of headlines this week. We just talked about two of them or also get they just had some huge updates with GitHub copilots. But here is one that I'm not maybe a fan of. So according to reports from TechRadar, Microsoft is reportedly considering a rebranding of its AI features under the name Windows Intelligence, potentially aligning its strategy with Apple's branding choices when Apple decided to call their AI Apple Intelligence. So again, this is just according to reports from TechRadar. So nothing official here yet, but references from an internal operating file suggest that AI features in Windows 11 may soon be grouped under a Windows Intelligence name, indicating a potentially strategic rebranding effort. So the name Windows Intelligence has been used by Microsoft in the context of cyber security. But this new application could signify a broader AI integration. By adopting a name similar to Apple's Apple Intelligence, Microsoft appears, according to reports, to be aiming to create a strong association between its operating system and AI capabilities in the mind of consumers. Let me just, I'm going to go ahead and insert my opinion right Normally I don't do this too much on our weekly AI news that matters shows. I don't think this is needed. I don't, I think the Apple Intelligence moniker is kind of of dumb if I'm being honest. You know, I don't think, you know, because Apple was so late to the party, you know, years years after Microsoft, years after Google, years after ChatGPT and Apple comes in and just, you know, they say, oh, this is Apple Intelligence, right? And for the most part, yes, it does do some things on device and it does so, okay, nothing great, right? The best feature of Apple Intelligence so far is its integration with, with ChatGPT. So I, I, I don't think that Microsoft needs to rebrand its AI offerings under a Windows Intelligence. Not a huge fan. I personally like the word Copilot. Right? That's where it's, it's, it's branded right? Now, I think that so many people out there are using this Copilot terminology to talk about or to describe an AI that you can work along with. However, Microsoft's. So I get that Microsoft's kind of in a weird position with this, right? Because everyone's using Copilot to reference in AI, right? But I would say the majority of people, when you talk about AI and Copilot, you're thinking Microsoft, right? So it's almost like a Kleenex, right? You don't tell someone, hey, pass me tissue, pass me unnamed. What? What? Right, you say Kleenex, whether it's a Kleenex or not. And I think that Microsoft has gotten to that point with its Copilot. So I'm not sure really why they would want to be moving away. Again, this is, you know, they probably have dozens of people working on this project who know way much, who know way more than I do. But not, not a huge fan. But the rebranding of Copilot, this isn't anything new, right? So we saw a complete rebrand if you're using Copilot on the web. So if you're using the free version of Copilot or the paid version, so not the biz chat, but otherwise you did see a completely new version of Microsoft Copilot, right? You saw rebranded, completely new interface with the new voices. So, yeah, we might see Microsoft continue in this direction. All right, AI agents at LinkedIn. Yeah, maybe. So LinkedIn has just launched an AI hiring a assistant that they're hoping will transform the recruiting process. So LinkedIn has introduced its first AI hiring assistant, aiming to revolutionize the recruiting landscape by automating up to 80% of traditional recruiting tasks, which is wild. So the new AI hiring assistant is designed to streamline various recruiting tasks, from creating job descriptions to conducting interviews, freeing up valuable time for recruiters. So the tool builds on LinkedIn's previous features, enhancing the capabilities of its generative AI to offer personalized suggestions through conversational memory so recruiters can input job descriptions, notes, and feedback, ensuring the AI's output aligns with their unique requirements. So concerns remain, though, that the increased reliance on AI could lead to bias in a lack of Personal touch in candidate interactions as seen in similar instances with other companies like Amazon. However, LinkedIn allows recruiters to selectively automate tasks, enabling them to maintain the human element essential in recruiting. So the other thing, I mean, this is great also, you know, pretty expected. LinkedIn has been pretty heavy on AI given it is owned by Microsoft. Right. However, don't fully understand this one either because unless they are going to end up charging a ton for this, right. If you're going to automate 80% of traditional recruiting tasks, isn't this also going to cut down their revenue a lot? I mean, I'm sure they've thought this through. But think, think of this. Let's say you have a team, let's say you're a big recruiting agency, right? And you have a team of a hundred recruiters, right? Is this only going to automate 80% of all of those recruiters jobs or might you only need 20 of them in theory, if this works out. So you know, I like the push here by LinkedIn because I think so many of these tasks in recruiting, right? I've spent so much time recruiting people to small, you know, small businesses that I own as well as companies that I've worked with in the past. So I understand that a lot of this is mundane and, and it should and can be automated. But to say that you can automate up to 80%, I mean, you have to think like, okay, is this either going to be terribly expensive, right, Once it's kind of fully rolled out, or are just, you know, might you have to, if you have a team of a hundred recruiters, maybe is your team going to be 50 or maybe is it only going to be 20? Right. So it's, it's, it's, it's kind of Weird, right, that LinkedIn is, is hoping to automate a big part of its marquee software, right? LinkedIn Premium LinkedIn Sales Navigator. Sales Navigator LinkedIn Recruiter. Right. So, yeah, interesting move. So we'll have to see how this one pans out once it fully comes to market. All right, our next news story for this week. China is leveraging Meta's llama model for military AI development, raising security concerns. So recent reports from Reuters indicate that Chinese researchers linked to the People's Liberation army have adapted Meta's llama model for military applications, highlighting the ongoing global competition in artificial intelligence. So a June paper from six researchers associated with the Academy, the Academy of Military Science, detailed the creation of, of chat bit or chat bit, I'm not sure which one it is. In AI, I think it's actually Chat Bit, an AI tool based on an early version of Meta's LLAMA model, specifically tailored for military intelligence and operational decision making. So Chat Bit reportedly outperformed other AI models, achieving capabilities comparable to OpenAI's ChatGPT4, although the researchers provided limited details on its performance metrics and operational status. Meta has stated that any military use of its open weights models by the PLA is unauthorized and contradicts its acceptable use policy, which prohibits applications in military warfare and espionage contexts. So despite these restrictions, Meta acknowledges the challenges of enforcing its policies due to the public availability of its models, which has led to concerns about the potential misuse of open source AI technology. Yeah, so I, I think we've kind of known that this has been going on for quite some time, but not specifically, right? I think anyone that, that reads a lot of AI news or you know, like myself, that gets to talk to very smart people, right? We all know that these open source models are being used for bad purposes. Sometimes there's exacts, sometimes there's not, right? So here this exclusive report from Reuters, a little scary, right, that China has been using this for these open models for military purposes. So this does raise some concerns on ways that you can better track how these models are being used and for what purposes. And I mean we might see in the same way that the US government has recently enforced some restrictions on companies exporting GPUs as an example, we might see some similar restrictions on, you know, exporting models. And I know that may not make sense now, right? But think back in the day, you know, downloading music, right? And you know, you could ban IPS or something like that. So you know, I wouldn't be surprised if we see a move from the US government sometime soon because as, as Meta's models specifically get better and better and better, you know, these closed and proprietary models, you have so much more control over knowing who's using them and who's not, right? Because in theory the companies know that, but at a certain point, right, when these open source models, they get forked, you know, they get fine tuned and re uploaded, right? So that is the nature of an open, you know, open source, open weights. I know Meta is not truly open source, but with an open weight model is, it just keeps being built upon, right? It's, it's almost like a cup of water, right? If, if you keep pouring it in 10 different glasses, pretty soon you're gonna, you know, forget where it came from, which is problematic, right? So that is the the downside and the ugly side of these models that are very open is it can end up in in the bad, like bad hands, hands that you don't want it into, such as US kind of enemy forces in these Chinese military operations. All right, and our last AI news story of the day. Didn't see this one coming, but Open AI CEO Sam Altman in a Reddit Ask Me Anything session discussed, well, just about everything. But one of the biggest thing here was discussing the GPT5 timeline in some future AI releases. So in a recent Reddit Ask me anything post, OpenAI CEO Sam Altman addressed the anticipated release of GPT5 and also provided insights on the company's ongoing struggles with compute capacity, which he said is affecting product timelines. So Altman clarified that there is no imminent release plan for GPT5, stating we have some very good releases coming later this year, but he emphasized that none of them will bear the GPT5 name. So yeah, a lot of speculation about what does that mean. You know, people are saying, oh, it could still be GPT5, but it could be this new Orion model, or it could be a GPT 4.5. But Sam Altman did essentially say there won't be a GPT5 this year. Which is something that I'd hate to be saying this, y'. All, I've been saying this for, for, for forever. I said there will not be a GPT5 in 2024. I said it would probably be first quarter of 2025 if we do actually get it. So the ongoing challenges in compute capacity are also a is a significant factor in delaying the rollout of these new AI models, altman said. And he also noted that as models become more complex, OpenAI faces some tough decisions about resource allocation. Right? They have all these other products that also require a lot of resources. But recent reports which we talked about in the newsletter last week, suggest that OpenAI has been working with Broadcom to develop a specialized AI chip to enhance its computational power, although this new technology may not be available until 2026, if it's available at all. All so some other tidbits from this Ask Me Anything session. So the advanced voice Mode feature for ChatGPT was initially expected to include vision capabilities, but those plans have been postponed again due to compute limitations. Right. So same Almond saying, hey, we don't have enough access to compute. So the rushed demo of that feature was during a press event, was intended to just draw attention away from Google's IOD developer conference, despite concerns about its readiness. So there we go again, you know, having these working demos really just being used as marketing against the other big company and you know, here we go. It's, it's. I mean who's actually shipping here? So also Altman mentioned that the image generation tool Dall E is currently lacking any release timelines for future versions. And the video generation tool Sora has also encountered some tech technical setbacks that hinder its competitive edge against other AI video systems like Luma and Runway. So Altman reiterated that improving the reasoning models right Now. So the O1 preview, the O1 mini and eventually the 01 full model are OpenAI's top priority with several exciting features on the horizon. All right, so yeah, pretty interesting. AKA again I'm on vacation so I was not expecting Sam Altman to do a tell all kind of Reddit ask me anything. So nothing there. Terribly shocking aside from him saying that the top priority was seemingly working on the reasoning models and that a GPT5 model would not be released this, this year, where we did have have previous reporting as recent as last week saying that a new model would potentially be landing in December after the US election. Right? So we've always said here at the election tomorrow, right? We've always said that there will not be. And by we, I've always said that there will not be, you know, a big Claude sonnet, uh, release, which there wasn't. Uh, I said there won't be a Google Gemini 2.0, which there wasn't. I said there won't be a GPT5.5 there. There wasn't until after the election. Even though every other person out there, you know, a year ago was saying, oh we're going to get these models. I'm like no we're not. Not until after the election. And then, you know, we kind of heard these December timelines, but it looks like at least for GPT5, we are not going to be getting it in 2024 and the priority is the reasoning models. All right, so that was a lot y'. All. Let me just do a very, very quick recap recap of the AI news that matters for this week, November 4th. So chat GBT introduced its enhanced web search capability. So not exactly the search GPT, but it does look like this is going to be the, you know, kind of the replacement for that very capable, very impressive next Google in response, or actually right before OpenAI did launch its real time search for Gemini AI in the API, setting the stage for some intense competition in AI powered search. Speaking of launches, Claude also launched its dedicated Mac app, its dictation features, and a new three tiered PDF processing system inside of Claude Google announced delayed for announced delays for its Project Astra and said it might not launch until 2025 at the the earliest. Speaking of delays, Microsoft also announced another round of delays for its copilot plus PCs ability for the recall feature, but it looks like now we make it that in December. Then we have Microsoft a report from TechRadar that said Microsoft may rebrand its AI features as Windows Intelligence. I hope not, but it could happen. Also, LinkedIn launched its AI hiring assistant. China is reportedly leveraging Meta's llama model for military AI development and we got a whole bunch of open AI and chat GPT news when Sam Altman did a Ask me anything on Reddit. All right, that was a lot. As it is every single week, y'. All. I hope this was helpful. I'm sorry that the audio and the video is not as good as it normally is, but we'll be back next week for more. So thank you for tuning in. And hey, shout Out Australia. I'm enjoying it here so far. If if you're listening on the podcast, if you're still listening, if you're from Australia, reach out. Would love to just, you know, give me a recommendation or two. So thank you all for tuning in. I hope to see you all tomorrow and every day for more Everyday AI. Thanks y'. All.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit your everyday AI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.