40 | Sam Altman is back as CEO of OpenAI, while Google Gemini is Delayed Again, and More AI News From This Week

0:00Hello and welcome to a short news edition of the Leveraging AI podcast. in today's show. We'll just review updates from this week.

0:13The first big news is obviously the return of Sam Altman as CEO of OpenAI and the return of Greg Brockman and a few other people who already said that they're leaving to start a new department that will do AI in Microsoft. If you somehow missed all the other big news about this topic, The nonprofit board of OpenAI decided to fire Sam Altman and demote Greg Brockman from being a board member to only just holding his president position. He resigned shortly after. A few other people left with him and more or less most of the employees threatened the board that if they will not bring Sam back, they will leave with them.

0:51Microsoft immediately hired Sam and Greg Brockman to come and work at a department that they're going to start for them as an AI division of Microsoft. And Microsoft also made an open offer to any scientist, anybody who works at OpenAI to come work for Microsoft for the same exact compensation. That obviously put a serious threat to the existence of OpenAI, which led to the fact but the board that was very clear that they're not wanting to get Sam Altman back to basically resign and let Sam take his position back and Greg as well. A new board is already chosen and that board is going to assign even a bigger board that will hopefully prevent that in the future.

1:34They also all agreed to give a third party the opportunity to investigate exactly what led to this situation so they can not have this craziness happen all over again. So as of right now, and this is being recorded, Sam Altman is the acting CEO again of OpenAI. Everything is back in its place. The only thing that is now questionable is the charter of OpenAI of putting humanity's best interests ahead of capitalism best interests, which was the charter. And now with everything that happened, I don't even know if that's true or not anymore. Only time will tell. The interesting thing that everybody wants to know beyond what exactly happened is the main assumption right now is that OpenAI made some kind of a quantum leap forward from a technological perspective in its research that has scared the board enough to say this is beyond the line where we're willing to go and release this to the world.

2:34And so I think it will be very interesting to learn what that thing is, but nobody is sharing that information, at least as of yet. So we have to be patient and see how that evolves. And from OpenAI to one of their biggest, at least potential competitors, Google, Google has been working on their next generative AI model called Gemini. Gemini was supposed to be released already, and now it's presumably has been delayed to Q1 of 2024. Gemini is supposed to be a multi-model model that has been developed from the ground up in order to compete with OpenAI and its partner, Microsoft. And my assumption is that after BARD was not impressive, to say the least, I think Google cannot allow themselves to release nothing short of spectacular.

3:26And so I think they're working very, very hard until they get to that point where they can release a model that is at least as good as GPT-4 and preferably better on at least a few things. And I think that is the main reason for the delay. It's obviously not a simple task to develop a model that competes with the best model in the world right now. But if there's a company that can do that, that's Google because they have more computing power, more data, more smart scientists, more resources and money to do these kind of things. So I have zero doubt in my mind that Google will play a very serious role in the future of humanity and AI in it.

4:10And so I think whether Gemini gets released in its new schedule in Q1 or whether it's delayed again, I think once it comes out, it's going to provide a serious competition to OpenAI and the best, most advanced generative AI model out there. Another piece of news from Google is that they have released the capability for BARD to review a YouTube video and provide you with the highlights as a summary in text. So you can take a link from YouTube, drop it into Bard, ask it what the video is all about, and it will give you a quick, short, and to-the-point summary, including defining what genre it is, what type of video it is, and potential follow-up questions that you can dive into for deeper understanding on the topics that the video has covered.

4:59I think what we'll see going back to the previous news from Google is that more and more integrated capabilities across all its different tools and software platforms, which I think will be hugely beneficial to us as users. just think about being able to use AI across everything, G Suite, YouTube, Google Search, et cetera, when it will know what happened in the previous steps across the other platforms, that could be amazingly powerful and obviously a benefit over what OpenAI has that just provides the large language model, but it will be somewhat similar to what Microsoft is doing with OpenAI's capabilities where they're integrating it into everything Microsoft, including Bing Search, and the 365 suite, as well as their cloud services.

5:44Moving from Google to another big player in the field, as you heard me say several times before, Meta has been playing a very different game than OpenAI and Google and Anthropic and so on. They have been mostly doing stuff for themselves, as well as developing and releasing open source models. And they've just released a very interesting model for image editing and video. They call it EMU. One is called EmuVideo and the other is called EmuEdit. EmuVideo is a text-to-video generator that is open source and available to everyone. And apparently, from what people are saying, it's pretty impressive.

6:22And per what they're saying and per human evaluation, it's performing better than any other text-to-video model that exists today. I haven't checked it myself, but it sounds very promising. You heard me say that before. I think 2024 is going to be the year of text to video. We've already seen some great attempts. The leader right now, I think, is Runway, but there's a bunch of other companies and a bunch of open source capabilities to do that. And now EmuVideo from Meta just provides some more capabilities to the mix. And EmuEdit allows you to edit images and create stickers from images and so on very easily just using text.

7:01And from Meta to Anthropic, Anthropic made two big pieces of news this past week. One is that they have released Claude 2.1, which is their next version of Claude 2, their large language model. I really like Claude personally. I use it a lot. The first thing that they've announced is that Claude 2 now is going to have a 200 ,000 tokens context window. That means that they're now back again as number one with the longest context window. They used to be in a very big lead over OpenAI, but AI in their release of GPT-4 Turbo announced that they're going to have a larger context window, 120 ,000 tokens over the 100 ,000 that Claude used to have.

7:44So now Claude has a 200 ,000 token context window. The difference between Claude 2 to ChatGPT-4 was really, really big and made a huge difference. The difference now is still big. It's 120 ,000 tokens for OpenAI and 200 ,000 for Anthropics Cloud, but I'm not sure how many people are actually going to use it. And the other thing I'm not sure is the level of accuracy of the responses you're going to get if you really upload something that big. But from a technical perspective, they are back in the lead. Another thing that this model does is they're saying it reduced hallucinations compared to the previous models by 50%.

8:26This is a huge improvement, but it doesn't mean anything because it's very hard for us to know how many hallucinations were there before and how much are there right now. And is it better or worse than other models like OpenAI and Google's? And Cloud 2.1 is available both on cloud.ai, the chat that you can do, that you can use online for free or a paid version. And it's also available through their API, which is really hard to get for some reason compared with getting the OpenAI API, which literally anybody can get. And the second big piece of news from Anthropic with regards to Cloud2 or Cloud2.1, I should say now, is that they have slashed their per token pricing dramatically.

9:09This is done, obviously, in order to A, compete with OpenAI that have reduced their pricing when they released ChatGPT for Turbo, but also with the release of more and more highly capable open source models, which are becoming more available and more stable, meaning large companies and enterprises can now have a real choice when they come to decide what models they want to use. They can go with open source models, which obviously are going to cost them very little to run. And so companies like OpenAI and Anthropic will have to find ways in order to make their services significantly more attractive financially.

9:49This is it for the news this week. And we'll come back with a regular episode with an interview on Tuesday morning, as always. And until then, explore AI, test new things, and share what you find with the world and with me on LinkedIn. And if you like this podcast, please give us a five-star review and share it with people that you know that helps us reach more people and helps us get you better guests. And that's it. I'll see you on Tuesday.