Everyday AI Podcast – An AI and ChatGPT Podcast - EP 429: AI News That Matters - December 30th, 2024

Episode Date: December 30, 2024

Google is using Claude to improve Gemini? Why is OpenAI looking at building humanoids? What does a $100 billion price tag have to do with AGI? AI news and big tech didn't take a holiday break. Get caught up with Everyday AI's AI News That Matters. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on AIUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:1. Impact of Large Language Models2. Google's AI Strategy3. OpenAI's Restructuring and Robotics Research4. AI Manipulation and Concerns5. AGI and its Valuation6. DeepSeek's Open-Source Model7. Meta's AI Plan for Social MediaTimestamps:00:00 Open-source AI competes with proprietary models.04:21 DeepSeek v3: Affordable, open-source model for innovators.07:39 Meta expands AI characters, faces safety risks.10:42 OpenAI restructuring as Public Benefit Corporation (PBC).17:04 Google compares models; Gemini flagged for safety.19:40 Models often use other models for evaluation.21:51 Google prioritizes Gemini AI for 2025 growth.26:29 Google's Gemini lagged behind in updates, ineffective.31:17 AI's intention economy forecasts, manipulates, sells intentions.35:13 Hinton warns AI could outsmart humans, urges regulation.39:24 Microsoft invested in OpenAI; AGI limits tech use.40:36 Microsoft revised AGI use agreement with OpenAI.Keywords:Large Language Models, Google’s AI Focus, Gemini language models, AI evaluation, OpenAI Robotics, AI Manipulation Study, Anticipatory AI, Artificial General Intelligence, DeepSeek, Open-source AI, B3 model, Meta, AI Characters, Social Media AI, OpenAI corporate restructuring, Public Benefit Corporation, AI investment, Anthropic's Claude AI, AI Compliance, AI safety, Synthetic Data, AI User Manipulation, Geoffrey Hinton, AI risks, AI regulation, AGI Valuation, Microsoft-OpenAI partnership, Intellectual property in AI, AGI Potential, Sam Altman.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

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. Meet Firefly AI Assistant, now live in Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. There's a new open source AI model leader.

Who is it? Why are Google workers reportedly using Anthropics Claude to train their models? The godfather of AI increased the likelihood AI will crush humanity. What does that mean? And why did Microsoft put a $100 billion price tag on artificial? general intelligence. Yeah, we might have had the holidays and still in the middle of the holidays craze here in the U.S., but AI did not sleep. So, at least this week, we won't either. What's going on, y'all? My name is Jordan Wilson and welcome to Everyday AI.

This is your daily podcast newsletter and podcast, wait, podcast, live stream newslet, we do so much. It's your daily free lives, three podcast and daily newsletter, helping us all not just keep up with what's going on in the world of generative AI, but how we can use it to get ahead to grow your company and your career. If that sounds like you, bet, welcome. You are in the right place. The other right place for you is your everyday AI.com. That is our website. So yeah, you can go sign up for the free daily newsletter where every single day we recap our live stream podcast. But also on there, there's more than 400. You can go back, listen to them all sorted by category. It is a free generative AI university. So make sure you go check that out. All right. So if this is your first time listening, maybe it's your 430th.

We do this almost every single week on Mondays going over the AI news that matters. So you don't have to spend hours every single day trying to keep up. You can just tune in on Mondays and we cut it all down for you. We break it down, cut through the marketing. we'll often tell you what matters. So with that, well, first of all, live stream audience. Hey, how's it doing?

Yeah, like what Jackie said, last few of 2024. Bronwyn, thanks for joining Richard and Samuel and Brian and Joe. Thank you, everyone, for tuning in. All right, let's get to the AI news that matters, y'all. A lot going on. A new AI model leader. Yes, at least when it comes to open source models

and it is extremely impressive. All right. So DeepSeek has launched its new open source V3 model competing directly with OpenAIs GPT40. Yes, we have an open source large language model that is competing head to head on benchmarks with the most powerful models, proprietary, closed ones. So the V3 model is now available on. GitHub allowing users around the world to utilize and modify the model, promoting collaboration and innovation without the constraints of proprietary software.

So yeah, if you're brand new to the large language model game, you're like, what's open source, what's closed source, and there's some things in between. But open sources, anyone can go and download them and run them, right? And costs are extremely low. So you can use deep seek online, more on that here in a second. So deep seek emphasized that V3 has achieved performance, comparable to top proprietary models like GPT-40 and Quad Sonnet 35 in various benchmarks, highlighting its capabilities and logical reasoning and problem solving.

So it uses a mixture of experts or M-O-E architecture and V3 optimizes its performance while remaining accessible to developers and researchers who may not have the resources of larger corporations. So this release comes at a time when demand for flexibility and customizable AI solutions are on the rise, attracting interest from companies and individuals looking to integrate advance AI into their projects without incurring a high cost. So deep seek positions V3 as not only a cost-effective alternative, but also a strong contender for organizations focused on developing specialized, applications and coding in mathematics. The model is governed by a license that permits reproduction, modification, and commercial

use enhancing its appeal among startups and independent developers looking to innovate without facing legal barriers. So pretty big news here out of from Deepseek and their new V3 model. It is a fraction of the cost of others. and it's open source, y'all. So this is from a Chinese company and, y'all, 2025 get used to a lot of AI news and large language models, news coming out of China. But it is a fraction of the cost to use deep seek.

So it's still a brand new model. I've barely had the chance to play with it. But hey, if you have, maybe you're in our live stream audience. are listening on the podcast, send me an email. Let me know. I mean, do you all want to know more about these, you know, models, these open source models, models from overseas, right? Here on the show, the only kind of open model,

and it's not true open source that we talk about a lot is Meta's Lama models, you know, not true open source, but, you know, kind of in between. You know, I don't know how much interest there is out there in these models. You can obviously use this on Hugging Face. You don't have to have a supercomputer in order to download this, but it's a fraction of the size, a fraction of the cost, will be dropping all of those bulletpoint details in the newsletter. All right.

Our next piece of AI News That Matters. This one's weird. I'm not personally a fan, but it's not going to impact me too much. But according to reports, Meta is expecting AI. to start posting on social media. Yes, they are wanting to flood their platform with AI users. Yeah, not real humans. They're just saying, yeah, we're going to have a bunch of essentially AI characters,

you know, posting on our platforms. Weird. All right. So according to the Financial Times, meta is set to introduce AI generated characters on Facebook, aiming to enhance user engagement through interactions. Similar to those with real humans. Yeah, are we living in a simulation? Yeah, every, you know, I don't know, our AI is just going to be, you know,

running their own social media. You know, you have one AI posting, a bunch of other AIs fighting with each other in the comments. I don't know. I'm not a huge fan, but META did state that these AI characters will eventually function like real human user accounts, complete with bios and profile pictures, the whole thing. So users will be able to create and interact with these AI characters through Meta's AI Studio, which has already gained some popularity for generating virtual relationships. Yikes. So the AI Studio currently features hundreds of thousands of characters and meta plans to expand

access to more countries over the next two years, focusing on making AI interactions more social. So this obviously raises a ton of safety risks, right, including the potential for exposed. using users to inappropriate content and obviously the spread of misinformation and disinformation. Also, there's not a ton of information yet out. This is just the one reports from the financial times, but there's not a lot of information yet on how meta might plan to police these AI generated users that are going to be out there running amok on their social media platform. So hopefully there will always be some sort of either watermark or something to let everyone else know that this is an AI user, whether it's commenting or posting stories.

So I do hope and assume that meta will at least go through that. To me, that is the bare minimum of what would actually have to happen in order for this to be successful. But I don't know. I don't personally see the value of this. It does look like, you know, people may be able to make within Meta's AI studio. You know, you could make a bunch of, you know, different personas and then potentially, you know, release those personas to go interact on their own behalf, right, to have agency on social media. So, you know, you might go assign some of your AI avatars to go interact with, I don't know,

real human content, other AI avatars. I'm not sure. Again, maybe this is, you know, I'm old and not cool. I don't see the point of this. You know, I think there's already, there's, it's already going to, it's already been hard enough over the past decade for people, especially on Facebook to understand what is real, what is not, especially when it comes to news, misinformation and disinformation. So then when you add in kind of this human, quote unquote, right, this human element

I'm not sure I'm a huge fan of it, right? Because I think people can go do their own, right? Especially, I think, you know, news that is not accurate, spreads pretty quickly on Facebook. But, you know, there's ways to verify that. There's not really great ways to verify who's real and who's not on social media. So like I said, hopefully, Mattel will be extremely strict in how, when, why these kind of AI personas can interact on their platforms. All right. Here's a big one. New information is out on Open AIs new for-profit structure. So Open AIs recent announcement, they did put a blog post on their website a couple of days, you know, here in the U.S.

after Christmas. So Open A.I's recent announcement to transform its structure into a public benefits corporation. So you probably, if you follow AI news, you're probably going to be hearing a lot about PBCs this week and next week. So they've announced their plans to transform their structure into a PBC, highlighting its urgent need for capital to support its ambitious AI research initiatives. So according to reports and OpenAI did release a blog post on their website, which we will be linking to in our newsletter. So Open AI is aiming to create a public benefit corporation to facilitate easier capital

raising, stating it needs to secure more funding than previously anticipated. So the restructuring follows a $6.6 billion funding round that valued OpenAI at $157 billion contingent on removing profit caps for investors within two years. So the non-profit arm, or what Open AI technically is right now, will still remain a significant interest in the PBC, ensuring it remains well-resourced despite the shift to a for-profit structure. So critics, including original Open AI co-founder, now competitor, Elon Musk and MetaP platforms. They are opposing this transition, raising concerns about prioritizing profit over public benefit. I don't know, y'all.

I don't think, you know, there's obviously a lot of legal drama that's been going on for more than a year between Elon Musk and, you know, Open AI. But, y'all, like, Open AI was originally in 20. 15 established as a non-profit, that's not what it is anymore, right? I think people are griping and complaining about this, right? Like, oh, open. There's no open this and open AI. They're driving profit. And y'all, like, what do you expect?

What do you expect? Right? I don't think the original founders really foresaw how quickly large language models would catch on and the need for them to even still, right, for them to still achieve their quote unquote non, you know, their nonprofit mission, which was to essentially better humanity with AGI and AI, right? You need to have money to do that. Many billions of dollars, potentially trillions of dollars of funding. So, you know, 10 years ago, I don't know if Open AI could look into it's Christopal and see, oh, yo, we might need trillions of dollars in five years, right?

I think even the most optimistic claims 10 years ago could not have really pinpointed where we're at today. So this move obviously aligns Open AI with everyone else. I don't know why people think Open AI needs its own set of rules just because it was originally set up as a nonprofit. But it's, y'all, as someone that worked at a nonprofit for 10 years, right, obviously not. this type of nonprofit, but it is very common, extremely common for nonprofits to either branch out and create something like a PBC or to essentially or eventually turn into a for-profit corporation

and keep the nonprofit a smaller piece, but something that can still help align its long-term goals. This is very common in the nonprofit world. So I don't know why this is turned into a year-long battle. But anyways, Open AI finally came out with some more information detailing their ambitions and goals around this PBC. So like I said, if you want to read that, we'll be linking to it in our newsletter. So Dr. Scott asking, any thoughts on the Musk lawsuit trying to block this? It's for show, right? Once this original lawsuit came out, I did like an hour-long podcast on that.

on that lawsuit. It's, I mean, again, I'm not a lawyer, right? It's, it's without merit. I think it's a joke. I would say most of what Elon Musk is putting out there in the lawsuit is really just, I think, to, how should I say this? He has created a direct competitor in XAI. Their large language model, GROC, is supposed to be upgrading from, you know, V2 to V3 here any day. I think Elon Musk wants to keep himself X-A-I and GROC in the headlines as much as humanly possible. He now has a position of power, right, in the U.S. federal government, which I'll hold my thoughts on that. That's wild to think about. But continue to, right, especially with that, especially with Elon Musk, having a, you know, unofficial, but kind of official position in the U.S. government. expect this to go on, expect this legal drama to keep going.

I would say, again, from a somewhat objective standpoint, it's without merit. I would say his lawsuit, I mean, go read it, judge for yourself. It doesn't take long. Even for fun, I put it into the lawsuit into every single large language model and just said, hey, is this thing have legs? Is this real? And every single large language model, essentially, right, aside from trying to to be politically correct about it, said, no, this is, you know, essentially a joke of a lawsuit.

All right. Here's an interesting one. And I have some personal insight on this. So Google contractors are reportedly using Anthropics Claude to improve their Google Gemini. All right. So according to reporting from TechCrunch, Google contractors are tasked with scoring Gemini's outputs, against Claude, focusing on criteria like truthfulness and verbiosity. Is that how you say it, verbosity, verbiosity, with up to 30 minutes allocated per prompt evaluation. So internal communications, again, this is according to reports from TechCrunch, so internal communications reveal that references to Claude are appearing in Google's evaluation platform,

suggesting a direct comparison between the two models. notably Claude's responses prioritize safety more than Gemini, often avoiding prompts deemed unsafed, whereas Gemini often is being flagged for safety violations in certain responses in the testing. So, inthropics terms of service, prohibit using Claude to train competing models without explicit permission, leading to speculation about whether Google has obtained such approval. A spokesperson from Google DeepMind confirmed that while model outputs are compared for evaluation, they do not train Gemini using Anthropics models. So, you know, kind of working in the gray area there.

And concerns have obviously been raised by Google contractors about their ability to accurately assess Gemini's responses on sensitive topics like healthcare, indicating potential risk in AI-generated information. All right. I'm going to go ahead and reveal something. I haven't revealed before. A trillion dollar company reached out to me and recruited me to help them train their models. All right. I'm not going to say who. Anyways, you know, just for fun, I went through the process just to learn more, right? It's not every day a trillion dollar company reaches out to you to say, hey, can you help us train our large language models? But from the process and seeing how these contractor agreements generally work with these large tech companies, I think this is happening all the time, right?

Even going back to our initial big story with Deepseek, the new Chinese large language model, the V3 model, right? A lot of people are talking about online. If you ask Deepseek, you know, what it is, you know, what are you based on? You get these responses from Deepseek that says, hey, I am GPT4. I am based off of OpenAI, right? So I'll probably do a dedicated show on this at some point, maybe in 2025. But I can tell you this, the combination between remote hybrid workers, you know, using contractors to help in various aspects of training large language models, all models.

and y'all, I'm not talking about Google, Claude, whatever here. All models use outputs from other models to evaluate, period. And I would go ahead and say this is, you know, I'm guessing a little bit here, but I will say it is very safe to say that certain large language model makers also use other large language models, specifically the GPT models to train their models as well. So I don't think this is necessarily, right, if you're someone that, you know, like myself, this is not shocking, but I think this is grabbing a lot of headlines right now, right? Like, oh, my gosh, you know, Google contractors are using Claude to either evaluate and maybe

train its Gemini models. The evaluation part, very common, right? I will go and say, I would say most models, right, when they're training their, sorry, most model makers when training their models, they're probably running the exact same prompt through at least eight different large language models to see how theirs compare, right? Especially when they're going through that reinforcement learning with human feedback, right? So the RLHF, right? And they're trying to tell a model, hey, when this input, when a user enters this input into the model, it should have, this type of output. This is an example of an output. This is a good one. This is a bad one. Guarantee everyone's using every other large language bottle out there, right, to see,

because you want to be safe. You want to make sure that a large language model has good, you know, strict guidelines. But obviously, like we just saw in this story, you know, claudanthropic, you know, specifically states that you cannot use their model to train. Right. So there is this ambiguity that I think a lot of companies operate in, training versus evaluating, right? So you're not necessarily using outputs to say, hey, this is how we should respond from our large language models, but evaluating, right? It's very common. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI assistant now live in the Adobe Firefly app, the all in one creative AI.

Studio. Powered by Adobe's creative agent, Firefly AI assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the assistant. The assistant orchestrates multi-step workflows, drawing on 60 plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any

time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adopi.com. Our next piece of AI news. Google is making Gemini its top priority in 2025. So according to reports, Google is intensifying its focus on artificial intelligence, particularly its Gemini language models as it aims for significant growth in the coming year.

So the company is working to establish Gemini as a leading AI product, targeting half a billion monthly users by 2025. So according to reports, during a recent, strategy meeting, Google CEO Sundar Puchai, emphasized the importance of Gemini stating that the stakes are high amid increasing competition in the AI landscape. So Google aims for Gemini to be the 16th product

in its portfolio to reach 500 million monthly users joining the ranks of popular services like Gmail and YouTube. So Google revealed plans for substantial updates to Gemini envisioning a universal AI assistant capable of handling various tasks across multiple devices and formats. Pachai acknowledged the rising prominence of rival systems, particularly open AI's chat GPT, which has become synonymous with AI for many users. So yeah, a big piece of this report of Google's kind of internal strategy meeting is essentially when people out in the world, everyday people talk about AI, people just automatically assume

you're talking about chat chippy T, right? So he's kind of voicing this concern or wanting to change the narrative that when people are talking AI and large language models, everyone automatically assumes they're talking about chat GBT, right? Because chat GPT was the first, right, the first big company to offer large language models on the front end that users could chat with that didn't require, right, like a back-end API or using a third-party service. it became synonymous with AI.

It exploded overnight. So Google executives are confident, according to the report, that Gemini will offer a richer and more integrated experience leveraging Google's extensive platform capabilities to outpace its competitors. So despite facing scrutiny from global regulators regarding potential monopolistic practices, but Chai stressed the need to focus on delivering innovative technologies responsibly. So Google employees, though, according to reports, express concerns about the potential costs associated with using their AI tools.

Essentially, they are saying, yo, we don't want a $200 per month plan like Open AI has. However, leadership assured them there were no plans for high subscription fees, such as that $200 per month. Here's the thing, y'all. I'm not going to say it's too little too late for Google. And I've been extremely harsh on Google, especially, you know, early 2023 and early 2024 because Google was asleep at the AI wheel.

They were. You know, it was almost a year. It was almost a year after ChatGBT, BT that Google released what was then called barred to the public. For a company like Google, they dropped. the bag. I think Google could have a stranglehold on the AI market. They have access to all the users. They completely failed in 2023 and early 2024. Let's be honest. I think they won the last quarter, right? Their December with the release of their experimental Gemini 2.0, Gemini 2.0 Flash,

right, those are the experimental, not the final versions. Their new research tool, all their agenic tools, a lot of them in beta. I think Google won the fourth quarter of 2024 because of their strong December release going head to head with open AI when Open AI had their 12 days of OpenAI. So I do think Google is now a serious competitor, at least, whereas if you would have asked me six months ago, and companies do ask me all the time. FYI, right? Because people always ask, yes, companies pay, pay me, pay us, pay everyday AI to consult them.

And I tell people, and I've told people six months ago, don't touch Google. Here's why. Google is fantastic. Google AI Studio, their vertex, all of their back end platforms have been great since they launched. Google's Gemini front end has been abandoned up until December. So you're right. When you log in to, you know, Gemini.com. and you're quote unquote chatting with Gemini,

it has been absolutely terrible up until December. That's because Google, whatever there, go-to-market strategy is with their large language models, was completely backwards, right? What you were using on Google Gemini was generally a model that was six to nine months behind their most recent model. So if you went into Google AI Studio, Google was competing all along with OpenAI.

head to head. What people don't understand, let me say this. This is free consulting advice for you people at Google, at Claude, everywhere, right? The people evaluating large language models are not always CSOs. They're not always CTOs. They're not always developers. They're not always software engineers. A lot of times for enterprise companies, it's someone like a chief marketing officer. It's CEOs. I've sat in the boardroom with C-suite of a company doing $20 billion in revenue. So an enormous company. And they're evaluating models with front-end platforms. Okay. Because this whole like democratizing AI, yes, that's happening. But guess what that means smart people at Google who fumbled the bag. That means you have non-technical people who are evaluating models for the largest

companies in the world on the front end, they're not logging into your Google AI studio as great as it is. They're going to Gemini.com. They're going to chat gbt.com. They're going to claw. AI. They're going to mistral. They're making those decisions because AI large language models has democratized access to AI. You have non-technical, non-technical people going in there to front-end products, not using the AI, not using developer councils. And they are making decisions. for multi-billion dollar, the largest companies in the world. Some of them aren't, but some of them have been evaluating models for the last two years that way.

So Google absolutely made a, I would say in the long run, trillion dollar mistake by doing that. All right. I'll get off my rant. Sorry, y'all. I know this is AI News that matters, but had to get a little rant in there. It's been a while. All right. Our next piece of AI News, open AI.

is apparently reconsidering a humanoid robots. All right. So according to reports from the information, Open AI is reportedly considering again developing its own line of humanoid robots. So the company has reportedly reestablished its robotics research group after closing it in 2021, indicating renewed interest in the field. Open AI's potential move into humanoid robotics could position it alongside industry leaders like Nvidia, Tesla, figures, and other, while also

highlighting its strategic investment in companies like Figure AI and OneX. So yeah, Open AI has been investing a lot of money in other robotics companies, but now, according to reports, has reestablished its robotics research group after closing it about three and a half years ago. So although humanoid robots is a lesser focus compared to Open AIs advanced AI models, these developments could significantly impact industries from warehouse operations to consumer applications. Yeah, there were other reports that we shared in our newsletter last week and the week before, also talking about how Nvidia is deepening its focus on robotics and humanoid robotics as well. And without giving too much of my AI predictions away for 2025,

I get used to talking about humanoids and robotics a ton. It is both exciting and extremely scary. I'm not going to lie to y'all. All right. Could AI be manipulating humans? Yeah, yeah, it is. All right. So a new study from the University of Cambridge reveals that artificial intelligence

tools may be used to influence online decision making, raising concerns about the potential consequences for consumer behavior and democratic processes. So researchers at Cambridge's Leverhume, I'm not going to get that right, Leverhume Center for the Future of Intelligence described in emerging intention economy where AI assistance can forecast and manipulate human intentions selling this information to companies. So this intention economy is seen as a successor to the attention economy that we've been

living in here for decades where social media platforms profit from capturing user attention. So now it is going into this AI predictive intention economy because AI can influence and manipulate humans, according to researchers. So the study suggests that large language models like chat GPT could be employed to anticipate user needs and steer human decision making based on psychological and behavioral data. So advertisers, could create personalized ads using generative AI tools, dramatically adjusting their strategies based on real-time user interactions and preferences. So the research indicates that tech executives foresee a future,

where AI models will accurately predict user intentions and present tailored information to guide their choices. So yes, we finally have a official study from a reputable institution confirming what we've already known to be true for a couple of years. So yes, AI can manipulate humans. Not only that, but large language models have already been used for a very long time in creating synthetic data, right? And guess what? That synthetic data or essentially, you know, these large language models are creating data that is used to train other models, right?

So model distillation. But also, it's using synthetic data for synthetic user testing, right? So big companies, instead of running focus groups with, you know, dozens or hundreds or thousands of humans, they're just using synthetic AI data. So this is not new. But I do think as we start transitioning from kind of 1.0 or step one models to step two. So going from, you know, kind of GPT models, transformers, your traditional chatbots to models that learn, models that. models that reason, models that self-improve, and models that ultimately can manipulate humans, right, and can kind of predict how humans will react, right?

So I think when you start to combine online user tracking with the data that all of these companies have on us, with large language models, yes. I don't think it's wild to say or wild to think about that AI will be able to intentionally manipulate people's opinions, right? It's already been, I think, in the darker parts of the web, been being used for a very long time. Hey, speaking of doom and gloom news, here's more. All right. So the godfather of AI, Jeffrey Hinton, has raised concerns over AIs now increased potential to threaten

humanity's future. So in a striking statement, Jeffrey Hinton, one of the most prominent figures in the history of AI has increased his estimate that the likelihood of AI leading to human extinction within the next 30 years has doubled. So initially, he said there was a 10% chance. And now the Nobel Prize winner in physics has said that there is a 10 to 20% chance that AI could lead to human extinction. Yay. Isn't that fun to talk about?

So in a recent interview, he expressed concerns that humanity may soon be outsmarted by AI systems, likening humans to toddlers in comparison to these advanced. technologies. So like I said, previously Hinton had estimated a 10% chance of catastrophic outcomes from AI, but now suggests that risk has doubled and has saying a 10 to 20% chance. So he highlighted the absence of historical examples where a more intelligent entity has been successfully controlled by a less intelligent one, raising alarms about the implications of creating artificial general intelligence or AGI. So Hinton, previously. worked at Google, resigned from Google specifically to speak more freely about the risk associated

with AI, emphasizing the potential for misuse by bad actors. He called for government regulation of AI development, arguing that relying solely on corporate profit motives will not ensure safety. And that's a great way to wrap up today's show, talking about company profit. Because at least according to reports, that is how Microsoft and OpenAI are defining AGI. All right. Yeah, a lot of buzzwords in the AI news this week. I know, but artificial general intelligence AGI is essentially when AI is way smarter and way

better than humans at almost any task, right? And I say this a lot. If you're looking at definitions of AGI from 20 years ago, we've definitely already achieved it, right? People with Open AI's new O3 model, which costs like a trillion dollars to, you know, ask one question. People are saying, oh, this new O3 model is AGI. I don't think so because today's definition of AGI is changing by the minute. But there is some important, I think, decisions that are being made right now between Open AI and Microsoft on defining what AGI even means. So,

according to reports, the two companies agreed to define AGI as a system capable of generating $100 billion in profits. So this is according to reporting from the information. So Open AI's public definition of AGI has been described as a, quote, a highly autonomous system that outperforms humans at most economically viable work. Sorry, economically valuable work, end quote. And I would say, even according to that definition, If you've used O1 Pro, right, even though O1 Pro, Open AI's reasoning model, it's most powerful, right?

So we had O1 preview and O1 Mini. Now we have O1 Regular and O1 Mini, but also if you are on a $200 monthly plan, you have access to O1 Pro. I have access to it. It is mind-bogglingly good, right? So it doesn't have access to all of Open AI's other tools right now, you know, the internet, the ability to upload spreadsheets, right? like using some of these other very important tools. It doesn't have access to all of those yet. But when it does, I don't think you could make an argument, right?

And as, you know, we've seen some rumors about, you know, some autonomous capabilities coming to chat GPT, like something called tasks where you can schedule different prompts to run at scheduled intervals, right? When you have that, when you can schedule a model like 01 pro to run at scheduled intervals and connect to the internet and handle with all your, and work with all your data. At that point, it's like, yeah, that's, that's AGI, right?

At least according to Open AIs definition, but that's not the definition anymore. So notably, according to these reports, any system classified as AGI will be excluded from intellectual property licenses and other commercial agreements between Microsoft and Open AI in their partnership, right? Because Open AI has reportedly received more than $13 billion in investments from Microsoft. And Microsoft reportedly has a 49% stake in Open AI. But there was this clause that essentially is like, yo, when we get, right, when Open AI gets to a GI, Microsoft can no longer use that technology in its. products because right now, even though we saw a lot of reporting recently that Microsoft is

diversifying, reportedly diversifying the models that you can use inside of its Microsoft 365 co-pilot platform. But right now, it exclusively runs on, as far as I know, it exclusively runs on Open AI's technology, right? So obviously Microsoft has hung its hat on the future of OpenAI's models. And there's been a lot of talk. recently, especially, you know, in the Microsoft circles is what happens when OpenAI says, hey, we've achieved AGI because previously it was Open AI's board of directors, their board members essentially, that said, hey, we've achieved AGI. And then at that point, Microsoft would only be able to use pre-AGI technology inside

co-pilot and its other platforms. And that's not a great future, especially if you are. a Microsoft fan, a Microsoft user. Right? So now, according to reports, they've changed that agreement because that is an extremely, you know, I'm surprised that it took it this long to get to where we at, to get to where we're at. Because to have a company like Microsoft, which, you know, depending on the day, is the

largest company in the world by market cap. To have it say, like, hey, board of directors, hey, handful of people at over, Open AI. You essentially get to decide when we've quote unquote achieved AGI. And then at that point, Microsoft co-pilot and its billions of users, right, of Microsoft systems can no longer benefit from the future technology that is considered AGI. Yeah, that's not a great way to determine, you know, the future of a trillion dollar company. So now it is essentially when a AI system has the track record or the capability to reach $100 billion in revenue. So interesting. Also, OpenAI CEO, Sam Altman has expressed optimism about achieving AGI sooner than expected,

suggesting its impact may be less significant than anticipated. Yeah, I think a lot of people are thinking, you know, oh, AGI is, it's going to be this like catastrophic line in the sand that, it's like, oh, after AGI happens, you know, the world completely changes. I don't think that's the, I don't think that's the case, right? Because it is going to be arbitrary, but now not so arbitrary, right? Before it was, you know, kind of like a vibe check, you know, Open AIs, Board of Directors and be like, yeah, yeah, we've achieved AGI right? Like I said, I think when an 01 model or a model that does chain of thought or reasoning

or a logic-based model, once it, you know, whether you want to talk about O3 or whatever, right, but once you give it access to everything that the GPT models have, right? If right now, if 01 Pro had access to everything GPT 4 has, right? So all the different, all the other, you know, chat GPT's canvas tool, its ability to browse, you know, chat GPT search, dolly, SORA, whatever, right? I know right now you can't access SORA within chat GPT, right? But if O1 Pro had all of these things and the ability to schedule prompt, or to schedule tasks, like, I don't know, I'd say that's probably AGI.

But now there's a $100 billion price tag on the AGI definition. All right. That's it, y'all. As a very quick recap, here are the AI news stories that matter for the week of December 30th, our last AI news that matters for 2024. So DeepSeek has launched their open source V3 model. and it is benchmarking very close to the world's strongest and most capable proprietary models, such as GPT4 and Claude Sonnet 3.5. Meta, yeah, this one's weird. Meta is expecting AI characters

to be posting on its social media platforms soon. OpenAI has released more details on its plan to restructure into a public benefit corporation. Google is reportedly, our, sorry, Google contractors are reportedly using Anthropics Clawed AI to compare and evaluate its Gemini responses. More Google news. Google is saying AI is its prime focus in 2025. About time, Google, you're only like three years too late. According to reports, OpenAI exploring humanoid robots. Researchers are saying that AI tools could manipulate online audiences, the godfather of a.

AI has says now there's a 10 to 20% chance that AI is going to extinguish all us humans. And Microsoft and OpenAI have reportedly put a $100 billion price tag on the definition of AGI. All right. I hope this was helpful. Y'all, if it was, please do a couple of things. Tell someone about it, right? If, you know, we, I think we only have one more show in 2024. If you gained, right, if you gain benefit from Open or from, you know,

Everyday AI this year, please tell someone about it. If you haven't already, go to Your EverydayAI.com. Sign up for our free daily newsletter. We will be recapping all of today's stories and a lot more. But also, please tell someone about this. If you're listening here on LinkedIn, a lot of work goes into Everyday AI, every single Monday through Friday. We'd appreciate it if you could share with your network.

tell someone about it, but also tell them to go to our website because there's like four, like there's so much thousands of hours of free AI information sorted by category, no matter what you care about by the world's leading experts. All right, y'all, thank you for tuning in to our last AI news that matters of the year. Hope to see back tomorrow and every day for more everyday AI. Thanks y'all. Meet Firefly AI assistant. Now live in Adobe Firefly, the Allman One,

A.I. Studio. Just describe what you want to create in your own words and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.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.