OpenAI’s Make or Break Lawsuit and the Golden Idol of AGI

00:05
Speaker 1
Ten. It's got the code. It's going to launch system. I know this. It's all the files of the whole park. It tells her everything. Sir, he's uploading the virus. Eagle one, the package is being delivered. You.

00:25
Speaker 2
Before we go too much farther, Sharon, can you introduce yourself and tell us who you work for and what you work on?

00:31
Speaker 1
Sure. My name is Sharon Goldman. I am a senior writer at Venture Beat, and I cover AI. I've been on that beat since April 2022. I started the first week that Dolly two came out, I remember, for the AI nerds out there, and it's been kind of a roller coaster ride ever since. I'd say it's sure to continue in 2024.

00:58
Speaker 2
You said you'd been covering it a while before that, too, and it was like more kind of back end. Can you describe what the beat was like before Chad, GPT, and Dolly and all this big stuff that we know now?

01:10
Speaker 1
Sure. I mean, I was covering it at a very high level. I was covering a lot of different emerging and digital technologies, but I feel like it was very much in the enterprise space, sort of embedded in applications that people were already using. I mean, AI had already been embedded in things like Google Maps and in know recommendation algorithms, like on Amazon, and then know enterprise applications, know predictive analytics and things like that. So those, I don't think, were things that were considered that sexy from a consumer standpoint. And generative AI really changed that Chat GPT, this kind of, like, user interface where you could actually ask something and get something in return, that was kind of a game changer for people to understand how they could use it in a very general way and not just under the hood, so to speak.

02:05
Speaker 1
So I think that kind of really changed the hype game and maybe why people thought it was a bit of a flash in the pan. But I do think it was kind of an expansion of use cases that were already happening. Like, there were already chat bots, but they were really bad. If you ever went to a website and tried to talk to one, they weren't very helpful, whereas today's chat bots, whatever else you might think of them, they're certainly more helpful. They can actually answer things if you prompt it correctly.

02:40
Speaker 2
What do you think about AI kind of becoming a stand in for? Or I guess AI kind of. When people say AI, I think they generally mean large language models, these generative AI things. What do you think about that kind of becoming a ubiquitous term? Do you think we need to disambiguate that?

02:59
Speaker 1
I think it's always been ambiguous. I think the word AI is a huge problem. I've spoken about this with people. I think the idea of artificial intelligence just immediately puts us in a mindset of like, science fiction, when we're really talking about math and statistics and patterns and analytics and things. Of course, generative AI takes us in a new direction where we're getting output from actual inputs, whereas before it was really just like algorithms that kind of found patterns in data. So perhaps if went back in time, we could have found a more mundane term for AI. But maybe it wouldn't sound as sexy either. But large language models are not the full breadth of AI at all. It's really one kind of branch off, I would say.

03:57
Speaker 1
But it's certainly a big branch off, as I say, because it allowed us to kind of talk to an AI through text or increasingly through images and soon audio and video, multimodal. And that input output is kind of like the big game changer, and that is becoming ubiquitous. And I don't think it's quite right to kind of bundle it all into this is AI. But that's kind of what's happening, at least from a consumer standpoint, and probably won't change anytime soon. It's kind of hard to change that jargon once it gets started.

04:40
Speaker 2
So the big story kind of kicking off. So we're here today to talk about a bunch of different things kind of looking at. We're going to be talking about regarding AI in 2024, which I think is going to be a lot. I think it's going to be one of the stories that dominates the year, sad to say, cyber listeners.

04:58
Speaker 1
Hope you like AI, because you're about to get a lot of it.

05:02
Speaker 2
Hope you like AI. Unfortunately, I was looking at the docket for like, the next three episodes, too, are all AI centered, which I also think is very telling.

05:13
Speaker 1
Yes.

05:15
Speaker 2
So we're going to have just as a little preview, we're going to talk about nukes in AI tomorrow with somebody, and then we're going to talk about a fiction book that is around AI. It'll be very interesting. But anyway, setting that aside, the big story kind of kicking off the year, or maybe it was right at the end of the year, actually, the end of 2023, is that the New York Times is suing OpenAI. What is going on here? What do they have to complain about? They're the New York Times.

05:51
Speaker 1
Well, I just want to say that I couldn't believe that this came out in between Christmas and New Year's. I was like, what a way to end 2023 with a bang. This is just how it's been all year long and will probably continue. It's actually, though, a story that I had been waiting on for months. I actually had written about this back in September. I had done this really deep dive into issues around copyright and the data that trains these large language models. I had spoken to someone who said that a looming legal battle between the New York Times and OpenAI was the big case to be waiting for. There have been other, many other copyright cases, lawsuits against OpenAI and Microsoft, and image generators like mid Journey and stable diffusion and Dolly.

06:54
Speaker 1
But a lawsuit between the New York Times and OpenAI, obviously the biggest newspaper arguably in the world and the biggest story around generative AI of 2023, OpenAI. A lawsuit between these two parties is big news and could potentially end up in the Supreme Court. And that's because the issue of copyright around the data that actually trained Chachi PT and other large language models is all around issues of fair use. Can these models scrape take content from across the Internet, including from the New York Times, and use it to train their models? And that's one thing. So they train the models with the content. What does that mean? That means that first of all, the output of the content, if the model's output content that is very similar to New York Times articles, it could be considered a competitor. So that's one issue.

08:05
Speaker 1
What if the output is not correct? We talked a lot about in 2023 AI hallucinations. So let's say you prompted something regarding the New York Times and the New York Times. The Chat GPT outputted content that they said, the New York Times said, but it wasn't true or it was a little off. That potentially could be seen as a reputation issue for the New York Times if their highly regarded journalism is now compromised in some way. So we're talking competition, we're talking reputation issues. And then, so the lawsuit that the New York Times filed against OpenAI and Microsoft, I should add, this was first of all, the first big media lawsuit around copyright against was also, I think it was only the second case that actually included Microsoft in the lawsuit, which is interesting also.

09:13
Speaker 1
So the New York Times is looking, they say that there has been billions of dollars in damages that they could potentially be owed. They haven't actually said a specific number, but it could be big. And for months I've had attorneys tell me that a copyright case around AI training data could and should come before the Supreme Court. That this is such a complex issue where both sides have some really good arguments. And the only way that this is going to be decided is to go before the Supreme Court. And the idea of such two big companies, this is the biggest case. This is the one that could really come before the justices. So that's why it's such a big deal.

10:04
Speaker 3
This is kind of a silly question, but I feel like when I think about a copyright case, I think about copyright over a piece of art or something that has already been produced. So here the issue is more with the data sets that are training the model. Correct. So you have the data set and copyright issues over that versus the output and potential hallucinations. So it feels like they're entirely two different know. I think, okay, this is like copyright law and this know tort, so to speak. That's very complicated.

10:40
Speaker 1
It is very complicated. I've learned a lot about it, but it remains very complicated. And that's exactly why it's going to end up for the Supreme Court. The lawyers I speak to say, because it's so complicated that both sides, you can really argue both sides. It's not a clear, the copyright laws are such that you can argue it either way. And that's why you also have people saying that copyright laws themselves should be changing, because this type of thing was never meant for the copyright laws of today. So the issue of the output is one thing. So that was a big deal regarding the image generators. So if you have a piece of art that the model was trained on, and then it outputted something that was either very similar in the same exact style, or you could prompt it based on a style.

11:35
Speaker 1
You could say, I want this person style, then wouldn't that be an issue for copyright? So that's one issue. When it comes to text, that has not been a strong argument so far. So, for example, there have been several lawsuits from authors around that issue. For example, Sarah Silverman, the comedian, she and other authors filed a lawsuit against OpenAI saying that in her case, her memoir was basically used by the model. It was trained, and you could actually ask it for different paragraphs from the memoir and it would spit them out. So that's a copyright issue in the way that you're saying. But the judge actually kind of threw that out, saying that it wasn't exact, and they didn't consider that to be a great argument. I'd have to look back as to why, but that ended up not being the strongest argument.

12:42
Speaker 1
But the issue of winding back even more, like unpeeling the onion even more. If you go back to the initial training of the model. So the models are trained initially on these massive corpora of data. And by massive I mean billions and billions, sometimes even trillions of what they call tokens, like pieces of text, not even whole words, like little bits here and there, but from all over the Internet from these massive data sets. And it's that training data that the New York Times doing is basically stolen. You've taken our copyrighted work and sort of vacuumed it up into your model. And whether or not it outputs the exact thing, that's not what they're arguing. They're arguing that the actual taking of that copyrighted work is illegal. And the fact that even if they're not outputting something, that's the same.

13:48
Speaker 1
When you chat with Chat GPT, for example, you might get little bits of information that then are not assigned to the article it came from. Or even if it is assigned, it might not be linked. And even if it is linked, people might not click on it. All of those things could affect the revenue of the New York Times too, because people aren't, they're not getting as much traffic, they're not getting as much revenue for subscriptions. So their focus is really on the actual training data that the models used.

14:21
Speaker 1
And they're saying that they tried to come to some negotiated settlement for OpenAI to pay retroactively for all the content, the New York Times content, that what they would say was stolen, but that they did not get an offer that they considered to be at the full value of the New York Times content, whereas OpenAI has negotiated with other media companies and come to that agreement. So for example, the Axel Springer, which owns Politico and Business Insider, they did come to an agreement with them to continue using their, you know, just as another example, Getty Images had a similar situation on the image generation side where they negotiated with OpenAI for their Getty Image portfolio. Yes, the output and the training are sort of two different things.

15:21
Speaker 2
We got numbers on that literally hours, either Bloomberg or I think Bloomberg reported it first, that the licensing deals are between one and $5 million to the news organizations for access to their articles for use as training data.

15:41
Speaker 1
I don't think they said that was the case for Axel Springer, which is larger, but I think they were saying that for some of the smaller media organizations that could be the case. I don't know who that would be, but I don't think they named names. And I think that article said that even for a small media organization, that would not necessarily be considered. The full value question is how much will other media organizations kind of come to their own decision to file a lawsuit and not go down the path of continuing to negotiate with OpenAI, or whether they'll just take the money and run as though this is the path that it's going down and they better get out while the getting's good.

16:35
Speaker 1
I think it really remains to be seen, but the New York Times OpenAI lawsuit has repercussions for media, but also for just the whole issue of how the models were trained initially. This article that I wrote in September, I kind of really dug into the fact that AI really is a research, academic, heavy discipline. It didn't really become an industry, so to speak, until the past five or ten years. So before that, researchers were using these big data sets to create models, but they weren't really thinking of it in terms of future products. They were really just trying to see how far they could go as far as science. So for them, these are academic researchers. They're not versed in copyright, they're not lawyers. They didn't hire lawyers. They were just using things with a research license.

17:38
Speaker 1
Creating these big data sets like, well, common crawl was not AI specific, but there are many other data sets that have been covered over the past year. Things like the pile books. Three was covered in the Washington Post several times. Big massive corporate data sets that were used to train these models. But they weren't initially trained with the thought that Chachi PT would come along. So they didn't really, even if they thought like, oh, this could be against copyright, they didn't really think there would be any repercussions about it because they didn't really think it would get to this point.

18:18
Speaker 1
But now that it's here, there's a lot of questions about the copyright issues around the data sets that train the models, as well as other issues in the data sets, whether it's bias, misinformation, other not safe for work, imagery and text and things like that. And unfortunately, creative workers have kind of become caught up in all of this. You have journalists whose work myself, there's a site that you can go to have ibintrained.com, I think is one of the sites you can actually look up and see whether your work has been used in these models. And I can easily see that my venture beat articles are in some of the training models, but you have artists that are realizing that all of their work is in the models. But in know, it goes beyond that.

19:19
Speaker 1
Social media, Twitter is now all of your Twitter x posts are being used to train Elon Musk's new XaI model. So we're all in there, basically.

19:33
Speaker 2
So I feel like, I instinctively understand for all the reasons that you just kind of laid out like, the New York Times side of this, right? As a journalist, as a writer, I do feel like my stuff is being stolen and used to train an AI model. But you said that this is complicated and it's big and that you can argue the other side. Can you argue the other side for me?

19:58
Speaker 1
So the other side is that the current copyright laws focus around the idea of fair use, and there are many ways that data can be used. Fair use. For example, you can use data for research. So if you had a research license to train a model, you can use data from across the Internet, you can use a lot of these big data sets. But fair use goes beyond that. There's been like a broad understanding that if the use of the data is for the common good, for example, or the common benefit to the society, there is some wiggle room when it comes to the use of data. Also, it's considered fair use a lot of the time if the output of the use of the data is not a one to one representation.

21:04
Speaker 1
So when we think of things as stolen, we're thinking of an actual thing that you're taking here. They're taking something, your content, and they're mushing it up, and then they're outputting something that might be similar but isn't exact. So if you look at a photo, if you look at a painting in a museum, a Picasso, and then you try and paint your own version, if you're skilled enough and it looks like the painting. Did you steal the painting? No, you didn't steal the painting. You didn't even copy the painting, especially if it's slightly different. So that's one argument, that the output is not exactly the same. Also, if you look at the technology itself, how is the output being created? So in the case of a large language model, it's not actually like taking the full book verbatim.

22:08
Speaker 1
It's breaking it up into tiny little pieces of data points and what they call tokens and predicting what comes next. So is that process of outputting, is that really stealing? Is it really like one to one coming out exactly the same? I feel like it's really easy to see this with the image generators, because you definitely can argue that a lot of the time it's not the exact same thing that comes out. You can put side by side, you can see that there are differences. Of course, that doesn't change the fact that for the creative worker, whether it's a journalist or a writer or an artist, the effect might be the same because you've created your worst competitor, your worst feared competitor who can be faster than you, that can output at a scale that you never could.

23:09
Speaker 1
So that's the other argument on the New York Times side or the creative side, that says one of the points of fair use that can be argued is whether it becomes a competitive thing. So the New York Times will argue that the output of Chat GPT actually directly competes with the New York Times journalism. But OpenAI could argue that it doesn't compete because it's not the same articles, it's not exactly the same. And that's where a lot of people are saying it's the copyright laws themselves that don't quite fall into what we're dealing with now.

23:57
Speaker 2
So let's say this goes through, and I know this is going to be a multi year adventure, I am sure OpenAI loses and they have to remove bits of trading data. Is that even feasible? What happens if New York Times wins?

24:17
Speaker 1
Yeah, so that's something that I've been asking people all year long also, is that if OpenAI loses, if these cases get to the point where the models have to be retrained, can they be retrained? The answer is no. You can't just retrain little bits of data. That's not going to work. You can't just take out the copyrighted part. You'd have to start from scratch. You'd have to throw away the original Chat GPT and start over. And I've had people say, well, so be it, start over. I have to say that it's really difficult for me to see how that would play out. And in fact, I kind of cynically wonder whether I'm not saying that this was in the plan all along, but the longer it takes to change the laws, the more all of this will be embedded in our daily lives.

25:17
Speaker 1
For example, Microsoft uses OpenAI's GPT technology to support its copilot that's now embedded in all of their office applications. So if you use word or Excel or any of their tools, Copilot is there to help you. And if those tools are based on OpenAI's GPT technology, how would you possibly unwind that without upending all of big tech? So it's really hard for me to imagine, but I've had people say, well, they should go retrain it. The other problem is that the reason OpenAI for example, when it comes to large language models, but also on the image side, if you look at mid journey or stable diffusion or dolly, the reason that they want the copyrighted stuff is not just because they are hoovering up everything, it's because the copyrighted stuff is the good stuff. It's the quality.

26:35
Speaker 1
You know, they're also training off of all sorts of crappy stuff, like Reddit posts and all sorts of total crap like Amazon review captions and things. But the good stuff is the quality journalism, the quality art. If you removed all of that, the models wouldn't work as well. And it's important to say that we don't really even understand why they work so well. So that's part of it, too. We don't really understand a lot of why the large language models work as they work, because OpenAI and other companies are becoming increasingly closed in their being proprietary about their research. So it's a little hard to tell what training data is affecting what part of the model. So that's part of where this gets interesting as well.

27:34
Speaker 1
But as far as how it plays out, I've spoken to attorneys and other experts who say, know it's possible that even if the Supreme Court ruled in favor of the New York Times, that wouldn't necessarily mean that the laws would change because Congress could then come forth and say, well, it's important for our society and our economy know this, AI continues. So we're going to pass a law becoming more liberal with copyright, for example. I'm just saying it could happen. For example, in Japan over the past year, they've made it very clear that copyrighted works are fair use for AI training. So a country can kind of just step in and say, well, this is just the way it is. But the New York Times is no dummy. They have really strong legal representation.

28:41
Speaker 1
I think one of their primary lawyers in this case is a firm that worked on the election machine disinformation during the 2020 election. And they also are representing the New York Times in another case where some of their best selling New York Times authors against OpenAI. So I feel like both sides have really strong legal representation, and that's going to make the arguments very strong on both sides.

29:18
Speaker 2
You kind of touched on another one of my questions, which was, how much will this matter in the long run when you have countries like Japan that don't have as strict copyright laws that have made it very clear that they are going to be using all of this stuff and consider it fair.

29:41
Speaker 1
Exactly. Yeah, that's exactly. I spoke to a lawyer for representing OpenAI in a couple of the cases, including the Sarah Silverman one, he would not speak on the record, but he did say that he kind of spoke to this in a way. This might be moot because companies like OpenAI could simply move their operations to another country, which is more lenient in their copyright laws, and go on their merry way. I guess he kind of felt like there's no putting this cat back in the bag. This is a global issue. It's not just a us copyright issue. And every country is different.

30:33
Speaker 3
Yeah, I'm just thinking, I, in preparing for this recording, I was, like, trying to remember, I don't know if it was hatchet or a different publishing agency dealing with. I don't know if it was a lawsuit or others coming forward. A lot of people have been complaining and being very understandably upset about AI using their materials to either train their models, et cetera, et like this New York Times case.

31:06
Speaker 1
Is it the first.

31:07
Speaker 3
It's the first major one that at least I can think of.

31:11
Speaker 1
It's the first major media company, yes.

31:15
Speaker 2
Are you talking about the Internet archive?

31:17
Speaker 3
Yes, that's exactly what I was thinking about. Different.

31:20
Speaker 2
The Internet Archive lost.

31:21
Speaker 1
Right. I was going to say that one case that a lot of people point to with regards to this issue is looking back at the big Google books copyright case that was ruled on in 2015. It was originally filed in 2005, so it went on for ten years through the appeals courts. And in that case, Google won, where they had created this project where they trained books from university libraries, millions of books made digital and searchable for anyone to use. And the authors Guild filed a lawsuit saying that it was against copyright. And in a way, it's very similar because you had creative workers who had authored books and were concerned that they would be affected by, you know, lo and behold, the appeals court ruled still in favor of Google that it was fair use. So that's an example. That's on OpenAI side.

32:33
Speaker 1
But there was another case recently that is more on the New York Times side, which was against. That was related to a photographer who sued Andy Warhol's. No, it was Andy Warhol's estate that sued a photographer who had sort of created a similar photo to the Andy Warhol original. But in this case, the photographer won. I think I'm sort of mixing things up here, but it was sort of like the other side.

33:10
Speaker 2
Yeah, I've got it here. Andy Warhol estate loses us Supreme Court copyright fight over prince paintings.

33:19
Speaker 1
It was a little bit of a different take. But it just shows the complexity of these copyright issues around fair use. And what lawyers have told me that what they're expecting is that different courts will rule differently and it'll depend on the case or different already. Like in the Sarah Silverman case, I think there were parts of it that have already been thrown out, but other parts that are moving forward. So the complexities are such that each case will be different, courts will rule differently, it'll kind of wind around. And ultimately at least one case they predicted would reach the Supreme Court, who would have to rule because the arguments are that complicated. Even if people take one side or the other very strongly, you know, and think it's very clear from a legal standpoint, it's not.

34:16
Speaker 2
So it sounds like AI is about to possibly change american copyright law, pull it into the new possibly.

34:26
Speaker 1
Possibly. Some people have said that this has happened other times when the VCR came along or when radio broadcasting began. But it does seem like in this case it puts such a new spin that perhaps it's time for the real fight for copyright laws to advance in one way or the other.

34:52
Speaker 2
All right, cyber listeners, want to pause there for a break? We'll be right back after this. All right, cyber listeners, welcome back. We're talking.

35:03
Speaker 4
Tired of ads barging into your favorite news podcasts? Good news. Ad free listening on Amazon Music is included with your prime membership. Just head to Amazon.com slash ad freenewspodcasts to catch up on the latest episodes without the ads. Enjoy. Thousands of Akash shows ad free for prime subscribers. Some shows may have ads.

35:23
Speaker 2
AI might be the most important new computer technology ever. It's storming every industry, and literally billions of dollars are being invested. So buckle up. The problem is that AI needs a lot of speed and processing power. So how do you compete without costs spiraling out of control? It's time to upgrade to the next generation of the cloud, Oracle cloud infrastructure, or OCI. OCI is a single platform for your infrastructure, database application development, and AI needs. OCI has four to eight times the bandwidth of other clouds, offers one consistent price instead of variable regional pricing. And of course, nobody does data better than Oracle. So now you can train your AI models at twice the speed in less than half the cost of other clouds.

36:06
Speaker 2
If you want to do more and spend less, like Uber eight by eight and databricks mosaic, take a free test drive of oci@oracle.com. Cyber. That's Oracle.com cyber. Oracle.com cyber.

36:20
Speaker 3
We're driven by the search for better, but when it comes to hiring, the best way to search for a candidate isn't a search at all. Don't search match with indeed if you need to hire, you need indeed is your matching and hiring platform with over 350,000,000 global monthly visitors, according to indeed data and a matching engine that helps you find quality candidates fast. Ditch the busy work, use indeed for scheduling, screening and messaging so you can connect with candidates faster, and indeed doesn't just help you hire faster. 93% of employers agree that indeed delivers the highest quality matches compared to other job sites, according to a recent indeed survey. When I was looking to hire someone, it was so slow and overwhelming. I wish I had used indeed leveraging over 140,000,000 qualifications and preferences every day.

37:10
Speaker 3
Indeed's matching engine is constantly learning from your preferences. So the more you use indeed, the better it gets. Join more than 3.5 million businesses worldwide that use indeed to hire great talent fast and listeners of this show will get a $75 sponsored job credit to get your jobs more visibility@indeed.com. Cyber just go to indeed Cyber right now and support our show by saying, you heard about indeed on this podcast. Indeed cyber terms and conditions apply. Need to hire you need indeed thinking.

37:42
Speaker 2
About the biggest AI stories coming in 2024. So can we move on and talk a little bit about the other big stories that you're tracking in 2024?

37:52
Speaker 1
Absolutely. Yeah. There's just a lot of really interesting things brewing.

38:00
Speaker 2
I think it's an election year that.

38:04
Speaker 1
I think is one of the biggest ones when I talk to people, and actually I just got off the phone with a company that is, their product is like an advertising technology product that allows you to clone voices, and they're actually jumping into the election game. They said that they would allow political campaigns to use their tool to like, say, if a, if a campaign is doing different things in a bunch of different locales, you know, they can swap out the clone voice to put in the new city or whatever.

38:46
Speaker 2
Nikki Haley is very busy and she doesn't need to record a discrete message to her constituents.

38:52
Speaker 3
I'm thinking of old concert advertisements on the radio where it's like, no, Metallica is coming to insert city.

39:01
Speaker 1
Here's a physical package being really that simple when you think about all the time it takes to personalize each little recording, I guess. But when I did ask about the, I mean, the big story around AI, though, is for real, serious misinformation. We've already seen it with videos being tweaked, using AI with Trump and Biden and even the pope. So I think we're all girding ourselves waiting for some really nasty AI disinformation videos or audios coming along. In this particular case, I want to make clear that they said that they would now not allow that kind of use, that they would have their hand in with any client using their tool for elections. But I do think that there's no doubt that it's going to kind of be awful.

40:06
Speaker 1
I spoke to one researcher who said that generative AI will make the 2024 us elections a hot mess, whether it's from chat bots or deep fakes. And there's been all this talk about regulating AI over the past year, and that might go by the wayside as different candidates kind of weigh in on how they feel about AI and how they're using it.

40:33
Speaker 2
I feel like that's one of those things that we tend to not regulate that kind of thing in this country until something really bad happens.

40:43
Speaker 1
Fortunate. And the problem here is that generative AI especially, is developing at such a rate that it's really hard to keep up. So to have all these long discussions about how to regulate keeps you kind of behind the eight ball. But on the other hand, you need to discuss it fully before implementing any regulation. So I feel like that's really difficult. In the EU, they worked to night and day to try and pass the EU AI act by the end of the year in 2023 because they knew elections were coming soon and the leadership would change and this was their opportunity. So here I do think that Congress has been hoping that they could get some regulations in the books before the fall of 2024. But I think that remains to be seen. As the election season really gets wound.

41:45
Speaker 2
Up, what would those regulations even look like? What can you even do?

41:50
Speaker 1
Well, there's been a lot of chatter during the fall. From September, October, November, the Senate was holding a bunch of different forums with all sorts of different leaders talking about how they could come to some kind of regulatory understanding around deep fakes, disinformation, election misinformation, safety, what has happened so far? In November, President Biden released an executive order using the only power he really does have to regulate AI at this time. And that would really be around agencies and the government. So putting different requirements for agencies as far as procurement or making sure that they are passing the test, as far as using safe AI tools and using that as a way to kind of set the bar for other companies to voluntarily basically sign on to do similar things.

43:02
Speaker 1
But that has its limits because it's not really a regulatory requirement, it's more of a suggestion or a hope.

43:12
Speaker 2
Well, I for 01:00 a.m. Excited for this election year. I'm sure it will be normal and fine and nothing bad will happen.

43:19
Speaker 1
In the meantime, though, I think we have several other big stories that will be fully in the news all the time in 2024. One of those is going to be OpenAI versus anthropic. Their valuations and funding have exponentially risen higher and higher over the past few months. Both of them are looking at new funding rounds and sort of touting what they claim is their current revenue from generative AI models.

43:53
Speaker 2
I think a lot of people know what OpenAI is, but tell us what anthropic is and what they do and why this fight is important.

44:00
Speaker 1
Anthropic is a similar lab, large language model developer. They are basically another OpenAI, and it actually spun out of OpenAI. So its founders worked at OpenAI and decided to go their own way because they felt that OpenAI was not, quote unquote, safe enough with their models, that they were not concerned enough about the potential risks of AI. It's important to note that anthropic, even more so than OpenAI, really comes out of the effective altruism movement. These are leaders who are very concerned that AI is going to destroy humanity at some point. So they really need to make sure to develop it, and develop it in what they consider to be a safe way. People have criticized that because the effective altruism movement is funded by a lot of billionaires, including previously Sam Bankman Fried, who's now in jail for his dealings in FTX.

45:06
Speaker 1
But Anthropic has its own chat bot, like Chat GPT Claude, and it's really gained a lot of momentum. They have a tremendous amount of funding. They started off big and now they claim that their valuation is going to be tens of billions of dollars at this point, and they're looking at another big funding round. So OpenAI, Anthropic has sort of positioned itself against OpenAI as far as the most sophisticated language models, the most powerful large language models, and also aligned them. Anthropic is aligned in the same way as OpenAI is funded, has Microsoft as its biggest investor. Anthropic has, among others, Google and Amazon as its biggest investors.

45:58
Speaker 3
I'm so confused about how effective altruism is involved in AI here, because that makes no sense to me.

46:08
Speaker 1
Well, effective altruism started more than ten years ago as a movement to be altruistic, more effectively how to do good better, and we can all agree with that. Over time, however, one of their biggest fears is AI. And they've become more and more concerned that AI is one day going to destroy humanity. An autonomous human like AI, in a science fiction way, is going to take over the world, or perhaps like be used technology used to start a new pandemic, have biotech weapons. So it's not to say that it's not still involved in other causes like climate change or animal rights, but it is highly focused on AI, which seems kind of counterintuitive, but they consider it to be their biggest priority now. And that's where it gets weird, because it's really been a ten year process.

47:23
Speaker 1
I'm planning on writing more about this soon, but it's been a very steady, steadfast process to invest in philanthropic projects to make sure that AI safety, they use the word safety not for kind of the current risks of AI, but what it could be in the future as far as taking over the world. And it's become such a billionaire funded landscape that it's really even infiltrated, like us government policy. So even Biden's AI executive order, there is some influence there from these folks who are wanting to make sure that they're on that wall, basically.

48:24
Speaker 3
Yeah. It seems to me the way that you're describing this is more like, let's make sure that AI doesn't do bad stuff versus let's have AI do good stuff.

48:34
Speaker 1
Yeah. And it's really interesting because every part, every use case of AI can be seen that way. There are two sides to any use case of AI. I was just thinking about that, because if you think of biotech, healthcare and drug development and things like that's one of the biggest touted positive use cases of AI. We all would love for AI to cure disease, to solve health care problems. But on the other side, you could say, well, what if AI takes over the world and starts a new pandemic or builds chemical weapons? So you really have those two sides of AI all the time, and it really depends on your angle of looking at it. And also, there's been a tremendous argument all year long about current AI risk. We were talking about things like election disinformation. That's something that's happening right now.

49:32
Speaker 1
Or are you going to focus on sort of this potential, very unlikely risk of an autonomous AI making us slaves?

49:44
Speaker 2
It speaks to the evils, may be reductive. The other ideology that is out there, that is not effective altruism, but is effective accelerationism.

49:59
Speaker 1
Yes. Which I find so fascinating because it really only developed as a name just over the past few months, even, definitely not even a year, I think in response to effective altruism and this focus on what we would call existential risk, this long term risk, possibly, of AI, you have the absolute opposite idea that, no, we shouldn't have any guardrails at all. We should be go develop at all costs. So Mark and know really signed on for that and became a big proponent and kind of hyped that up. But you have many other people adding ACC to their handles on X Twitter.

50:50
Speaker 1
What I thought was super interesting, if we remember Mark Schrelli from the drug pricing scandal, he's super into this and does all kind of Twitter spaces on it and stuff, but it's really just taking effective altruism, I think, to sort of this very opposite extreme. I'd love to see more kind of a balanced middle. That's what I try to promote when I talk to people. But you can see why it's happening because people are kind of taking sides. And that's something else I really covered a lot last year that I think is going to continue is sort of the politicization of AI, tribalness. It's really turning into politics. And I think it's just because everything's so uncertain. Ever since Chat GPT came out, I think we've all kind of been like, what is going to happen?

51:49
Speaker 1
Whether it comes to legal issues, how it's going to affect the workplace, everything about it is kind of uncertain. And in that uncertainty, we're kind of like taking sides and making our own predictions.

52:05
Speaker 2
Can we then get even more esoteric? You wrote something on your substac I thought was really fascinating, that the AI folks are criticized, especially people that are rushing to build an artificial general intelligence are criticized for wanting to create a God. Right. And I've certainly looked at some of the things that Sam Altman has said and been like, it sounds like you're trying to make a God in the absence of one. It's a little strange. Your argument is, no, they are not. They are, in fact, making idols. Idol? Yes. What is the distinction?

52:44
Speaker 1
Well, I think, first of all, you can't create a God like either you believe one already exists or not, I guess is what I'm saying. But in the absence of a God, if you have a group of people that their belief system has been shattered in some way, they're looking for something new, they're seeing a great deal of uncertainty in their world. What do they do? They create something godlike to worship, to bow down to hope for and root for. When we think of the biblical golden calf is what I wrote in the. You're. These are people who were worried. They worried Moses wasn't coming back, so they really wanted something to replace the God that they had been promised, and so they built their own.

53:38
Speaker 1
So here I do feel like there's a sense of we have the power even to build our own God, our own idol that is godlike, that is cult like in some way. So that's a little bit of a fear of mine that mindset kind of, like slants it in a way that isn't very helpful. But there's definitely those kind of belief systems out there, and I do think it sort of rises out of this sense of uncertainty. We all want control, and we all want to feel like we're in control. And if you feel like you have that power and you can build your own idol that is godlike, and you can get other people to buy into that, then it is a little scary, I think. And I often just sort of pick at that a little bit like that.

54:31
Speaker 1
Sam Altman and his ilk sort of sound a little bit like know. They're kind of, like, out there saying very confidently what they think is going to happen. But I really think that none of us know exactly what's going to happen. I mean, everyone's making bets, and that's fine, but we don't really know how it's going to play out.

54:52
Speaker 2
And you've got one of the people that's on the board burning graven images in the office and talking about op Chat GPT as if it were an entity of some kind, then we're already pretty much there.

55:12
Speaker 1
Yeah. I think as humans, we're just so prone to anthropomorphize things. I mean, don't you anthropomorphize your cat? I do. So I anthropomorphize my plants and my car. So I think it's no surprise that we're all out there kind of with a little bit acting like Chachi. PT is a real entity, a real living thing, but that is far from the truth, at least yet. So I do hope that we keep that realism in mind and maybe that would go back to if there were a different word for AI than artificial intelligence, it might help, but we're still out there using the word AI.

55:59
Speaker 2
I like word calculator.

56:01
Speaker 1
Yeah, there you go. I mean, that would just sound not as powerful at all, wouldn't it?

56:06
Speaker 2
I know, but I like it. I think Corey Doctro told us that was word calculator. I think I like that more.

56:14
Speaker 1
Yeah.

56:14
Speaker 2
It strips it of some of its power.

56:16
Speaker 1
Exactly.

56:18
Speaker 2
Here's a final question, Emily, unless you've got something else.

56:21
Speaker 1
No, please go ahead.

56:22
Speaker 2
I'll ask you what I ask everybody when we get on this subject. Do you use it and what do you use it for?

56:28
Speaker 1
Sure. People always ask me that, and I definitely do use it. But I do feel like I don't use it in the way people might think I would. I definitely don't use it to write my articles because I don't find it very helpful. I don't know if it's because it just isn't there yet or because it's never going to be. But I do use image generators. I use it for my stories. People have criticized Venturebeat for using dolly and midjourney. I do use it. That's been an incredibly useful thing for my pieces. I'm not saying I don't have mixed feelings about it, but I am using it. I'll just say that out loud. As for text, I use it sometimes to help with headlines. Also not that helpful. Not as much as you would think.

57:26
Speaker 1
Sometimes it's a starting point, sometimes in a very targeted sentence level. If I have a sentence that is very generic and I'm trying to think of another word or another way to say it, I might try it and go from there. The thing I use AI for most, maybe you guys do too, is otter is transcription. That's changed my journalism life. I mean, real time AI generated transcription has been a game changer.

58:00
Speaker 3
I know I forget that it's an AI, even though it's in the name.

58:04
Speaker 1
When I think of how I used to type while I talked on interviews for years and got carpal tunnel syndrome, that's been the game for me.

58:15
Speaker 2
It took a job that would take literally, like 8 hours, right. And made it instantaneous.

58:23
Speaker 1
And I used to pay. That's something I used to pay a lot of money for, too.

58:28
Speaker 2
Yeah, it's funny. I forget too, Emily. I've completely shut out of my mind the era when you would set a day aside and all you're going to do is transcription or you're going to pay somebody.

58:41
Speaker 1
I do think that's where I feel like the use cases for a lot of these tools will become very personal, depending on your role and just what you like to do. I don't think it's as straightforward as people think.

58:58
Speaker 2
Well, Sharon, thank you so much for coming on to cyber and walking us through this. I'm looking forward to all the craziness that we're going to be experiencing in 2024 with regards to AI, and we'd love to have you come back on and walk us through when something horrible happens.

59:13
Speaker 1
Sounds good. Anytime.

01:00:44
Speaker 4
Tired of ads interrupting your gripping investigations? Good news. Ad free listening on Amazon Music is included with your prime membership. Ads shouldn't be the scariest thing about true crime. Just head to Amazon.com ad freetrucrime to catch up on the latest episodes without the ads. Enjoy. Thousands of Acad shows ad free for prime subscribers. Some shows may have ads.