In this episode of Leveraging AI, Isar Meitis talks about diverse array of AI news of the week including the latest from Apple's AI developments to the fierce battle for AI talent among tech giants.
News discussed:
- More details about apples AI expected release
- Google just hired a former OpenAI executive to boost its AI efforts
- Perplexity’s growth upends SEO fears, reveals crack in Google’s dominance
- Sam Altman is trying to convince Hollywood that Sora won't destroy the movie business
- ChatGPT no longer requires an account — but there’s a catch
- OpenAI’s voice cloning AI model only needs a 15-second sample to work
- OpenAI and Microsoft reportedly planning $100 billion datacenter project for an AI supercomputer
- Microsoft Copilot has been banned for use by US House staff members, at least for now
- Meet Devika: An Open-Source AI Software Engineer that Aims to be a Competitive Alternative to Devin by Cognition AI
- Elon Musk's updated Grok AI claims to be better at coding and math
- Mistral AI Releases Mistral 7B v0.2: A Groundbreaking Open-Source Language Model
- Amazon Gives Anthropic $2.75 Billion So It Can Spend It On AWS Xpus
- Google's DeepMind CEO says the massive funds flowing into AI bring with it loads of hype and a fair share of grifting
- OpenAI removes Sam Altman's ownership of its Startup Fund
- OpenAI just released a Sora generated music video — and it’s like something out of a fever dream
Don't miss out on what's latest in the world of AI. Stay informed, stay ahead.
About Leveraging AI
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
In this episode of Leveraging AI, Isar Meitis talks about diverse array of AI news of the week including the latest from Apple's AI developments to the fierce battle for AI talent among tech giants.
News discussed:
- More details about apples AI expected release
- Google just hired a former OpenAI executive to boost its AI efforts
- Perplexity’s growth upends SEO fears, reveals crack in Google’s dominance
- Sam Altman is trying to convince Hollywood that Sora won't destroy the movie business
- ChatGPT no longer requires an account — but there’s a catch
- OpenAI’s voice cloning AI model only needs a 15-second sample to work
- OpenAI and Microsoft reportedly planning $100 billion datacenter project for an AI supercomputer
- Microsoft Copilot has been banned for use by US House staff members, at least for now
- Meet Devika: An Open-Source AI Software Engineer that Aims to be a Competitive Alternative to Devin by Cognition AI
- Elon Musk's updated Grok AI claims to be better at coding and math
- Mistral AI Releases Mistral 7B v0.2: A Groundbreaking Open-Source Language Model
- Amazon Gives Anthropic $2.75 Billion So It Can Spend It On AWS Xpus
- Google's DeepMind CEO says the massive funds flowing into AI bring with it loads of hype and a fair share of grifting
- OpenAI removes Sam Altman's ownership of its Startup Fund
- OpenAI just released a Sora generated music video — and it’s like something out of a fever dream
Don't miss out on what's latest in the world of AI. Stay informed, stay ahead.
About Leveraging AI
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
Hello, and welcome to a weekend news episode of Leveraging AI, the podcast that shares practical ethical ways to leverage AI to improve efficiency, grow your business and advance your career. This is Isar Matis, your host, and we have every week, an explosion of really interesting and really important AI news. So let's get started. We are going to start by additional details that were released about the expected Apple announcement about AI. As I mentioned in the last two weeks, there's been growing proof that Apple is going to release big AI related news in its coming worldwide developer conference. And this past week, researchers from Apple have released some additional information about what they're going to release. So they're calling the new family of AI models we"reference resolution as language modeling", also spelled as realm with a mix of lowercase and uppercase letters, and it will enhance serious capabilities to allow it to understand the context of what you're doing, the conversation you're having with it, and even the content that is on your screen. According to the benchmarks that were shared by these researchers, the performance is at least as good as GPT 4 while they're claiming that their smallest model out of the three models, they're presumably are going to release is at par with GPT 4 and other leading models and their bigger models are even better. And now I'm quoting from the researchers paper, we demonstrate a large improvements over an existing system with similar functionality across different types of references. With our smallest model obtaining absolute gains over 5 percent for on screen reference. We also benchmark against 4 with our smallest model achieving performance compatible to that of GPT 4 and our larger models substantially outperforming it. As we need to assume, this has been in the making for a while because I don't think anybody assumes that Apple has not been working on this since the launch of chat GPT. So they've had a year and a half to work on these models. Like everything with Apple, they don't release details until they actually do, and then it's usually as part of a big announcement and the release of new devices and a new iOS platform. So within a couple of months we'll get to experiment and actually see what it's doing. But it sounds very promising. And not surprisingly, it's gonna be tied in into Siri that will allow you to operate everything you need to operate within your Apple ecosystem. That being said, as I mentioned before, more and more companies are coming up with models that are quote unquote, better than GPT four, which was the most advanced model in the world until recently. But chat GPT 5 is coming. GPT 4 has been released to red teaming over a year and a half ago. So it just showing you how good it was if companies are just catching up to it 5 comes, it's probably going to be a game changer. That being said, As I mentioned, Apple is saying that on their bigger model, they're substantially passing what GPT 4 can do. So game on. Now, I want to jump to a topic we started talking about last week, which is the movement and even poaching of AI talent across the leading companies in the world. So this week it was reported that Mark Zuckerberg, personally has been writing emails to leading AI researchers at Google deep mind to recruit them to meta. In these emails, Zuckerberg emphasizes the importance of AI to meta and the future, and he's trying to tempt them by coming over by offering them higher salaries, better term, the promise that by the end of 2024, Meta is going to have over 340, 000 NVIDIA H100 GPUs. And obviously the fact that they are pursuing the open source world versus the closed source with which a lot of developers are attracted. Zuckerberg also mentioned to investors that Meta's largest investment in 2024 will be obviously AI engineering and AI compute resources, which makes perfect sense. But that's just Meta. In a news from Google, they announced they just hired Logan Kilpatrick, which was the head of developer relation at OpenAI, and he was brought over to Google to lead the development of its AI studio and Gemini, which is Google's AI chatbot. At OpenAI, Kilpatrick was the one that quote unquote helped bring their products to life. and this kind of move at that kind of level shows you how much is companies are fighting amongst each other for talent. If you remember just last week, we shared that Microsoft was hiring most of inflections, AI leading staff. And we also know that Salesforce CEO, Mark Benioff was publicly pitching a open AI researchers to join Salesforce. So the party is on. The amounts of money that AI talent is currently getting is absolutely insane with numbers going up to a million dollar salary for hiring leadership from smaller startup, bringing them into the companies like Meta and Google and the bigger players. This war between the giants is obviously just good competition. It will make it harder and harder for smaller startups to actually compete in this space. In addition to all the stuff we talked about last week, when just the access to more data, more compute, and more resources will make it very hard for smaller companies to compete. This is just another aspect of the same problem where if you were able to put together a team of really good AI talent, they will most likely be poached by bigger companies with deeper pockets to come and work for them. But every now and then a company is able to defy these rules, at least for now. And one of these companies is Perplexity. We talked about Perplexity many times on the show. I love Perplexity. I use them all the time. I use them more and more over just using Google. And Perplexity is as if Chachapiti and Google had a beautiful baby. That allows you to enjoy the benefits of a large language model together with the benefits of a research engine. According to bright edges research, perplexity referral traffic to brands website has grown 40 percent month over month since January of 2024, which is showing an explosive growth. Now, the fact they're growing 40 percent doesn't still means they're putting a big dent in Google's traffic, but it's definitely showing that the capability to have a better user experience than the old school type of search we have known is definitely coming. Another interesting piece of information from that research is saying that 60 percent of Perplexity's citations match to the top 10 percent of organic results on Google, which is indicating that it's getting to. Somewhat similar results, but that means also that the other 40 percent of traffic is not coming from that. That also obviously means that the other 40 percent is not aligned with Google's top results, which means that there's going to be some dramatic changes in SEO results. What I mean by that is if you're currently on the top 10 results in SEO, and your business depends on it. This may not be the case moving forward because there's going to be a lot, a much bigger variety of results. There's also, I believe is going to be significant decline in clicks to the companies who are resourcing that information. And I'm not saying that based on research, I'm just saying that based on my personal usage. As I mentioned, I'm using perplexity more and more in researching things. And in many cases, I don't click on the references because the information I'm getting from the summarized information of the AI summary is just good enough for my needs. And so I would say I'm probably clicking on the citations, maybe 25 to 30 percent of the time, which means. Two thirds of the time, I am not following the leads to the actual source of the information. That means that these companies are giving data to this engine, but is not enjoying the traffic. What does that mean? It means that if your company depends on SEO, you better a start learning how to feed this new mechanism so that your citation show up and B, you have to figure out New sources of traffic because SEO is going to be dramatically different in the probably relatively near future. In another interesting piece of news, Sam Altman, the CEO of OpenAI, has been meeting with Hollywood executives to discuss Sora, the company's new video generation AI tool. Now, I mentioned Sora as it came out in February. It absolutely blew my mind and the mind of anybody who has seen it. It generates extremely realistic videos in high resolution up to a minute long from very short frames. Prompts and that's just what they've released. I don't even know what they actually have behind the scenes. They're talking about the ability to edit the video as well as just generating it. So it's going to be a complete game changer compared to what we know today. So Altman and Bride Lightcap, who is the CEO of OpenAI, recently held meetings with the likes of Paramount, Universal, Warner Brothers, et cetera. And during these meetings, OpenAI asked studios executives to help with the rolling out of Sora. And some studios were receptive to using these tools in productions, suggesting it's going to save a lot of money. Now, if you remember, just last year, there was a huge long strike of actors and writers on Hollywood, fighting for their future. And part of the agreement that they were able to reach with the big studios was that they're not going to replace them with AI. And back then, if you go back on this podcast and listen at that time, you will hear that. What I said is that it doesn't stand a chance. In reality, the fact that they have these agreements in place will make absolutely no difference because once somebody else will be able to produce videos that are as good as Hollywood's without the same cost of production because of the amount of agreements they have in place with writers and with actors, And on production, then these new types of studios that will use AI will literally drive the existing large studios out of business, unless they do the same thing. So I don't see a good outcome of this situation. I think we'll have two types of movies. I think we'll have movies where human actors play the main role. And I think we'll have a lot more movies where there's not going to be a single human actor or even maybe writer. In the process, there's going to be a lot of other production efforts that will be generated out of this people who will know how to use AI in the best way. And obviously editors are going to be there and all the other stuff on promotion and things like that. But I think as a whole, this industry is going to take a very big hit in the very near future because otherwise it will not be able to survive. By the way, this past Monday, so I released the first third party videos produced on the tool. Meaning the first videos that were not produced by open AI themselves and their employees, and it's showcasing some amazing capabilities. So I suggest you go and check this out. And obviously they're not the only people in the game. So you have companies like runway and meta email video and Google's Lumiere. Which are also working on text to video capabilities and have been for a while, and they will probably close the gap on Sora as well. So even if Sora is not the tool that will replace actors and video production, at least in some capacity, other tools probably will. And staying on the topics of OpenAI, OpenAI just announced that you don't need a ChatGPT account anymore to try out ChatGPT. you can now use ChatGPT without signing up and without providing any login information, just with more limited capabilities. As an example, you will not be able to save your chat or use custom instructions and there's going to be a shorter context window as well as more restrictive content policies, but you will be able to use ChachiPT without giving it any additional information about who you are or even creating a login. This is obviously a move to get a bigger audience moving to ChachiPT as the competition is becoming more and more fierce. And if you listened last week, I was joking. We had almost no news about OpenAI. this week we have a lot. So OpenAI has just started offering limited access to its text to voice generation platform called Voice Engine, which can generate synthetic voices based on only 15 seconds clip of someone's voice. Several different leading companies are already using this engine in the back end, companies like Age of Learning, Haygen, Dimagi, LiveVox and Lifespan are all using this model behind the scenes. and there are obviously other companies that have been developing similar engine themselves, companies like, Eleven Labs and PodCastle that are offering similar tools. Now this on one hand is a very powerful capability that allows creators to create content very easily. And practically no editing requirements or even no recording requirements of any voice of every person in their company. That being said, it also opens the door to deep fakes and as a result, FCC recently banned robo calls using AI voices after multiple people reported that they received spam calls from AI clone voice of president Joe Biden. Now, OpenAI's partners that I mentioned before, agreed to abide by the usage policies that OpenAI has put in place that is aligned with not allowing to impersonate people or organizations without clear consent of them, getting explicit informed consent from the original owners. Voice speaker and disclosing to listeners that the voice are AI generated. While this is a great move in the right direction to reduce the risk, I don't see that as an actual full bulletproof solution. And I think, as I mentioned many times on this podcast, that we're moving into an era where we will not be able to know. What is true and what is not on digital communication, which is a big part of our communication. So any voice you hear, any video going back to our previous topic of video generation may or may not be true. And this could include your loved ones, your boss, et cetera. We will not be able to know. So verify everything. Every information you consume online to check if it's true, how will we verify as an example, go to large known sources and look at several of them to verify that the information you're looking at is accurate and correct. That's obviously not going to solve the problem because the vast majority of the world is not going to do this. And they're going to see something on their favorite social media and share it with the rest of the people on other social medias and fake news and fake rumors are going to spread like wildfire, way more extreme than we've ever seen before. Staying on the topic of OpenAI, but now about their relationship with Microsoft. Microsoft and OpenAI has several reports are sharing that OpenAI and Microsoft are discussing an investment of a hundred billion dollars in an AI supercomputer data center or data centers. The project is called Stargate and the goal is to have an AI focused supercomputer featuring Millions of GPUs. So earlier I told you that Mark Zuckerberg is planning to have 340, 000 of these GPUs by the end of this year, that means that by the end of this year, they're going to have more GPUs than probably anybody else on the planet, but now we're hearing that Microsoft and OpenAI are planning millions of these GPUs in the data center. That is going to cost in excess of 115 billion to put together, and the plan is to put this in the US, the final location of the data center or data centers has not been finalized, but the goal is to do it in several different steps, potentially four different systems that are going to be deployed between now and the end of 2028, When the first capacity most likely coming as soon as 2026. Now, this is not a huge surprise. I think these companies, it's obvious from all the actions of these companies in the past few years that they're see that they see their future and the future of computing in these data centers. But the downside of this that we mentioned several times in the past. One is that it will require gigabytes of power, and the rumors are saying that they're looking into various alternative sources of power, such as a nuclear facility that will feed just this data center or data centers, and the other problem is obviously cooling. This will require a huge amount of cooling capability because it will require a lot more computing power to a given space compared to existing data centers. Otherwise, it's going to be gigantic in size, which will cause problems. Cause other issues. And so these kind of solutions are things that humanity will have to come up with, hopefully without a huge impact on the environment. And since we started speaking about Microsoft U. S. House of Representatives, the U. S. House of Representatives Chief Administrative Officer, Catherine Spindor, issued a guidance stating that all co pilot features will be blocked and removed from all window devices used by the House of Representatives staff and members. And this is because it represents a potential threat of leaking house data to non house approved cloud services. Now this obviously should be the concern of every single company who is starting to use that information. If you have any proprietary data, Classified data or any kind of data that should not be shared with anybody other than the organization itself. You should be aware of the issues with these platforms And beyond the fact that there needs to be some it actions in place, this goes back to educating the people in your company, in your organization about those risks, because even if you block them on company computers, people are still going to use chat GPT on their phones. They're still going to use chat GPT on their home laptops that they also do work on, et cetera, et cetera. And so education has to be a very important part to make people aware of the risk of potential data leakage that may arise from using some of these tools. Now, Microsoft themselves has announced that they plan to launch a version of their AI chatbot designed specifically for government work. This summer, meaning they're trying to address the problem and make sure that the government can comfortably use these tools within a more secured environment. So I do think that solutions are on the very near horizon, but that may not be the case for your organization. And hence again, education is critical. And switching gears to a completely different topic, just in the last two weeks, I told you about Devin, which is a platform released by Cognition AI that allows to do very sophisticated development of code using AI beyond just writing snippets of code. It actually goes and research the internet and does debugging and solves problem and can also deploy code that was generated by the platform. So this week, a new tool was announced. It's called the Vika and it's basically the same thing as Devin only it's an open source platform. It is highly capable of comprehending high level human instructions, conducting research, and breaking it into smaller tasks, while being to execute these tasks across multiple aspects of code generation. And just like DevIn, it can gather relevant information from the internet, access project files, online resources, providing targeted recommendations, and insights that it can implement itself or be used by human developers. It is currently already showing impressive performance metrics, including high success rates in completing sophisticated coding tasks, correctly the first time around, Which is obviously leading to significant time savings for developers and development projects. I mentioned that several times in the past. The world of writing code is going to be one of the first ones to be dramatically impacted by these large language models and tools like Devika and Devin are definitely showing much more aggressive moves in that direction. The reason for that is they live in a very constrained world. If you think about large language models in general, they have to mimic the universe while the universe is very unpredictable and has a huge varieties of gray and very little black and white. It's exactly the opposite when you're writing code, when you're writing code, it's a very structured universe with very clear set of rules with a lot of black and white and very little gray. And hence these tools can mimic that world much more accurately, quicker than anything else. Another big piece of news in new platforms being released is Elon Musk's x. ai has announced that they're releasing a new version of grok, which is their large language model. This new version is called grok 1. 5 That has improved capabilities and reasoning as well as the ability to process up to 128, 000 tokens Which is aligned with what chat gpt4 is doing right now as far as the context window. Those of you who don't know what context window is, context window basically means how much data can one chat include without it stops being effective. So right now, 128, 000 tokens is what grok 1. 5 will be able to do, which is aligned with what chat GPT 4 can do right now. Now, according to the data released by XAI themselves, This model has a huge improvement over GROK version 1. Math benchmark score increased by 50%. The GSM 8K, which is math word problem score, increased by 90%. and human inval, which is a coding score, increased by 74%. These scores put GROK 1. 5 within the same area of performance As some of the leading models like GPT 4, Jebedi Pro 1. 5 and Cloud 3, which is a big improvement over the previous version of Grok. I personally didn't get in touch to test it because to get access to it, you need to be a premium member on X, which I am not. But as people start releasing actual benchmark and test cases, we will be able to know how good it actually performs. In parallel to this, there has been leaked information that investors close to Elon Musk are talking to help X. AI raise 3 billion to continue its AI development efforts in evaluation. And that 3 billion investment is going to put the valuation of X. AI at 18 at$18 billion. Speaking of investments, Amazon has exercised its option to invest additional funds into Anthropic. So the new investment is$2.75 billion. And probably most of it is in access to AWS GPU's cloud infrastructure. While this is new money coming into Anthropic, it is money that is a part of an older agreement to invest up to 4 billion. So the first part of it of 1. 25 billion was invested in the past, and now Amazon has committed to the rest of the amount. But this shows you this synergetic relationship that we are seeing across the board right now with the big cloud computing companies such as Microsoft and Amazon striking deals with the large language model developers, such as Anthropic and OpenAI, in this relationship, obviously, the computing power comes from the cloud infrastructure, and in return, they're getting access to the most advanced models that they can provide to their customers, but that again comes to show how hard it is going to be for smaller companies to compete in this world. Just think about the amounts that we were talking about 13 billion of investment in Microsoft already in open AI, 4 billion of investment by Amazon into Anthropic, a hundred billion dollars to develop new data centers by Microsoft. And if you don't have these kinds of relationships and you don't have this kind of access to compute power, it will just be very difficult. Very hard to compete And since we're talking about the impact of the big names, one of the big personalities in the field is obviously Demis Hassabis, the CEO and co founder of DeepMind, which is a company that Demis has founded years ago, and that was purchased by Google. And they're now a subsidiary of Google. So Demis was just awarded a knighthood status in the UK for services to artificial intelligence. This comes seven years after Hassabis has received the award of CBE commander of the order of the British empire for his services to science and technology. So he's Definitely one of the leading minds in the world today and has been for a very long time when it comes to AI. And in an interview this week, Demis has expressed his concern about the hype and the bubble surrounding the massive influx of funding coming into the AI industry. Based on PitchBook data, investors has poured almost$30 billion into 691 generative deals in 2023, and that's not stopping in 2024. Now, Demis is stressing. That we have to be realistic about the current capabilities and limitations of AI systems as they are today. And he's obviously raising concerns over the ROI of some of these investments. I will add that in addition to that problem that Demis raises, there's another huge problem. And again, it goes back to the big versus the small companies, many of these AI tools that are being developed right now to do something very specific, very well are going to become a feature within the capabilities of the bigger models, completely eliminating and evaporating entire ecosystems and companies that have gotten huge investments. As an example, we just talked this episode about the AI voice engine coming from OpenAI. There are several different companies, such as Eleven Labs that has worked very hard to develop that capability. If OpenAI is going to provide this option for free or for 20 bucks a month, and it's just going to be a part of the ecosystem of OpenAI. So for 20 bucks a month, I'm getting this capability. Everything, including the AI engine, I will most likely not going to pay another 20 to 11 labs to use their service. So my gut feeling tells me that the billions of dollars of investment money are going to evaporate in the near future from that exact thing that I just said. That's it for this week's news. We'll be back on Tuesday with a fascinating interview. The two interviews that we're releasing this coming Tuesday and the next Tuesday, you do not want to miss. They're both with really brilliant people. This coming Tuesday is with Matt Lewis. He's the first chief AI officer I ever interviewed. And the following week, it's going to be with Reed Robinson, the head of AI product in Zapier, both are absolutely fascinating interviews that are going to give you a lot of insight on what companies can and should be doing about AI implementation right now. My last request is if you have been enjoying and finding value in this podcast, please share it with other people. Literally put your phone right now and hit the share button, share it with a few people, that can benefit from it. And while you're at it, write a review and rate this podcast on your favorite podcasting platform that helps us a lot in reaching new people that needs this content and needs to learn about this stuff just as much as you do. So I would really appreciate it if you do it right now, unless you're driving, then do it once you make your next stop. That's it for now. Have an amazing rest of your weekend and I'll see you again on Tuesday.