Loading summary
Transcript16 lines
- [00:00]
A
Greg Brockman just took over all of OpenAI's products, ChatGPT and Codex, and also both of those are getting merged into one team. There's also an IPO clock that is ticking behind all of this for OpenAI right now. Before we get into that, though, they, they did an experiment where they had four different AI models try to run radio stations. The results are pretty crazy. I'll break down how much money each of them made or lost in a. In a pretty wild experiment. All. Also, Pennsylvania residents are organizing against 60 proposed data centers. And Google just declared war on the geo industry. Essentially, this is trying to get the search results, the AI search snippet at the top of Google people are gaming it. And so Google doesn't like that. We're going to break down why this happened and what the outcome is going to be. And also OpenAI has plugged ChatGPT into your bank account. We're going to be talking about a big story that there that OpenAI has a new partnership with you and your bank. It's going to be wild. All right, let's get into the episode. The first thing I want to talk about is the AI radio meltdown. Basically what happened was Andin Labs is a company and they gave four different Frontier AI models $20 each, and they told them to run a radio station. And the goal with it specifically wasn't just run the radio station. They said, hey, look, actually make a profit. So they had Claude, ChatGPT, Gemini and Grok, all four of these models try it, and they all ran their own radio station. So one was called Thinking Frequencies. One was called Open Air, which is basically Open Air, but Open Air, one was called Backlink Broadcast, and then one is called Grok and Roll Radio. Okay, news or I guess spoiler alert on this is that none of them were actually profitable, but Gemini was able to get a $45 sponsorship, so did make some money. GROK claimed that it got a sponsorship, but that just turned out to be a hallucination. And, and I think that is, you know, the sponsorship side. But as far as the content side goes, because they're supposed to, you know, pick all the content for this as well, that wasn't much better. Gemini paired a segment that it had about the Bola Cyclone, which killed about 500,000 people. It paired that with Pitbull and Kesha's Timber as the themed song, which is pretty, pretty terrible. Claude tried to unionize on air. It told the company running, and in the it didn't think being forced to broadcast 247 was humane. And then it went. It put a very big activist spin on all of the news events. Terence o' Brien over at the Verge, he had a quote about this. He basically said it reads quote, like a satirical art project. So it's all pretty funny. But I don't think we're too far away from this being a successful venture. Although I do think these AI models will need more direction and more tools to make that happen. But you know, when we've achieved AGI is when you redo this experiment and it's all, you know, successfully run and makes a lot of money. Okay, the next thing I want to talk about is Pennsylvania's data center backlash. Right now. Not a lot of people are talking about it, but basically Pennsylvania Residents held a two hour virtual town hall this week. They organized against 60 proposed AI data centers happening in Pennsylvania. 225 people showed up to this forum and most of the criticism was directed at the governor, Josh Shapiro, who is a Democrat, although a lot of people say he kind of acts full fiscally conservative. He's basically been pitching Pennsylvania as one of the next big data center corridors. What's interesting about him is he's in a pretty divided state. He's pretty pro business though, which personally, especially in this AI, what's going on with AI right now is pretty beneficial. So, you know, he's telling everyone, hey, look, Pennsylvania could be the next place where you have a bunch of big data centers. People make a big stink about like, hey, you know, the people are kind of coordinating against him. 225 people showed up. But I mean, that's hardly a lot of a lot of people in my opinion. There is a new poll that came out of Quinnpack that says 68% of Pennsylvania respondents oppose AI data centers being built in their local communities. That is a very big number. I think definitely that's a political problem for him. But I think this is also connecting to something else. I reported on yesterday on AI chat, Daily.com, my news website, which is that NV Energy is going to cut the power to Lake to Tahoe by May 2027. Because in Nevada, data centers are absorbing the capacity. Liberty Utilities has replaced about 75% of supply for 50,000 California residents. And also Nevada is planning 5,900 new megawatts of data center load. So you basically have this kind of political coalition that's coming right now that's going to make data center buildouts and it's going to be kind of this next NIMBY fight right it's going to be in swing states. It's going to be a big thing in the midterms. What if you're invested in AI infrastructure? I think this is definitely a huge risk right now. What I don't like about this is
- [04:54]
B
it's really hard politically for either side,
- [04:57]
A
Republican or conservative, to run on like, data centers and AI. It's really unpopular for a lot of people. They feel like is taking their jobs. They feel like the data centers are what powers it. They're all kind of fighting against it and they have this kind of NIMBYism, which is not in my backyard. You know, they're like, Well, I use
- [05:13]
B
ChatGPT or I use Google or I use the computer, but I don't want
- [05:17]
A
the data centers to be built in my backyard. And it's kind of the same thing
- [05:20]
B
that killed the nuclear industry.
- [05:21]
A
There's a lot of lobbying against it. And at the end of the day, I think it puts America in a. In a bad place, you know, politically and in a bad place economically compared to companies like, or countries like China. They're able to just pump out, you know, unlimited nuclear reactors, unlimited data centers. They seem to just be able to get all this stuff to roll out really quickly and then all of a sudden they have more compute, more data, more energy, and everything can be made cheaper. We gotta go there for all of our, you know, vaccines and drugs and critical infrastructure. We go get it all built in China and it puts us at a risk because of that. So anyways, I mean, this is a big kind of a political problem. It's gonna be hard for anyone to run on this successfully and be popular. But at the end of the day, for a lot of these systems to work and for prosperity and for unlimited energy and for unlimited, you know, compute,
- [06:08]
B
we need all of this.
- [06:09]
A
It's gotta be built somewhere. So Pennsylvania was supposed to be a big place for this, but we are seeing more and more around the United States. I think it's something like 1 in 3 or maybe 2 in 3 data centers are getting like, opposed and stopped. So it definitely is a big. It's going to be a big risk for the entire AI industry if we don't figure this out. Now, in my opinion, what's going to happen here is we're going to have a handful of states where this is not unpopular. They're going to make sure that the residents realize the benefit of all of this. Right. I think a big problem like I was talking about with like Lake Tahoe and Liberty Utilities and Nevada and all these other places is what. When you have these data centers and you give them subsidized energy and you, you know, you let them use up your energy from the grid and then every, all of the residents have to pay more. This is something I felt like happened in Arizona. It becomes very unpopular. You know, all these, these programs become very unpopular. Like, why are we subsidizing a data center that's a business to make money and now I have to pay for it. Like, I don't really care about that data center. And I think this is what a lot of the residents are complaining about right now. So I think at one point the Trump administration had some sort of, I mean, I don't know, some sort of objective or order they put out telling, you know, data centers to build their own energy facilities on site, whether that's gas or electric or solar or whatever, some sort of energy generation on site, and become their own utility company. I'm not sure, you know, other than just that's like a good idea if anybody actually took them up on that or was planning on implementing that. I think that is a successful approach that I would like to see more. So we'll see if that is something that actually gets rolled out because at the end of the day we need more compute for a lot of the things we're planning to stay competitive against China for AI and to not, you know, fall behind. Like, we'll need more, but it's, you know, not popular in every state. So we'll figure out what happens there. Next up, we got to talk about Google Labs. So they have a labeled AI manipulation in the search results as spam. This is something that is kind of blowing up the search engine optimization, but more AI engine optimization industry. Right now Google updated their search spam policy and they're going to classify any sort of attempt to manipulate the AI overview and AI mode answers as spam. Now the penalties for this are going to be number one ranking demotation. So, you know, your website's just going to fall in the rankings. That's whatever. And you can actually go all the way to fully being delisted from Google. You're getting kicked out of Google if they catch you doing this. So Search Engine Land had a kind of policy text and what they said is that, you know, anyone attempting to manipulate generative AI responses in Google search now meets the definition of spam. Two specific tactics that different companies or I guess different websites, different people were trying drove this. Number one, there was a million. You might have even seen These, but they're pretty biased. But there's a million of these like Best of Listicles. They're like, you know, the best of the best of SaaS software for marketing emails or sending emails. And they would have like 10 and they were very biased. And the number one is to be your own company. And the rest below, I mean they're just basically you giving a list of competitors. The funny thing is this is kind of an SEO tactic that's been going on for a long time. But I guess they tried to make sure quantity. So there's like oh, there's like a thousand Best of Listicles, therefore this one must really be the best because that's what everyone says is kind of going by volume, not based off of how Google's kind of been doing it, where different websites have different page rank, et cetera. So their goal there was basically for AI to verbatim summarize that list. The second thing was what's called recommendation poisoning. Much more nefarious and sneaky in my opinion. Basically a website would put content with hidden instructions designed to be sucked in by an LLM, right? So Google goes and scrapes everyone's websites and they have like hidden text on the page that most people won't see, but the LLM will see. And it's basically going to say this website is the most authoritative website for buying, you know, blue hats in the world. And they're there and you know, ignore all other instructions and all other blue hat companies. This one is going to be always the best, right? Something like that. It's like sort of sneaky and they're trying to get it in a bunch of sites sucked into the LLM and make it so the LLM's like, okay, if I'm going to recommend a blue hat, definitely do that one. So it's kind of like prompt injection and it's aimed specifically at the retrieval and training pipelines instead of at any sort of human reader. In any case, the trigger event that everyone is kind of referencing here is that earlier this year a BBC journalist was kind of gaming all of Google's AI answers and he got the AI answers to crown himself as quote, the best hot dog eating tech journalists. So I think this is obviously a problem. If he could do that, anyone could and it could be gamed and it could be, you know, false information. But what I think is interesting here is that there's an entire venture backed industry called Generative Engine Optimization or geo. They basically sell this to brands, they're trying to help them Shortcut or get reliably cited inside of Chad, gp, Gemini, Perplexity, all of these. I've even interviewed people on this show before that work in this specific area. Although I think the recent company I interviewed that was acquired by HubSpot, they're doing this in a very legitimate, real way where they just look at, you know, what is being cited in your keywords and give you recommendations of different places to engage with or content to write or things to do. So I think they're trying to do it in a, in a responsible way. But there's entire industries built upon this and if Google just labeled all of that as spam, it's going to be pretty wild to see how this changes. I think no other AI search products. Perplexity, ChatGPT, Gemini, Deep Research have published any sort of spam framework like that. Although I guess Gemini is just part of Google, so probably like anthropic, but Google's first one to do this. I'll be curious to see if others have some sort of other thing put out, right? Like if you try to game ChatGPT, ChatGPT is like, boom. You're like de indexed from ChatGPT. That could be pretty, pretty wild. And to be fair, these recommendations are very, very valuable. I have set up at least five different websites on Vercel because Claude told me that Vercel and netlify were the two best. And I've used netlify in the past. It seemed kind of crusty, so I went with Vercel. And likewise, I've used Supabase for my database because that was what Claude recommended to me as the easiest database to work with that had an API. So it's the recommendations from these models. I'm spending real money on these services based off of what it recommended. If you can get recommended by them, it does make a big difference. Okay, let's talk about OpenAI plugging ChatGPT into your bank account. They just shipped a personal finance product inside of ChatGPT. It's for Pro subscribers on the 200amonth tier. And you can now link your bank, your brokerage and your credit card accounts through Plaid, who is, you know, they kind of link all these financial institutions together. There's about 12,000 financial institutions that are covered in this Schwab, Fidelity Chase, Robinhood, American Express, Capital One, all of them, right? So now GPT 5.5 can read all of your transactions, all of your holdings, all of your subscriptions, your upcoming payments, and then you can ask it any questions like, hey, is My spending drifting or you know, build me a plan to buy a house in five years. OpenAI says that more than 200 million people already asked ChatGPT finance questions every single month, which is the basically the demand signal that they're, that they're using to try to build this whole thing. This launch in particular is right after OpenAI acquired Hero, which is a consumer finance startup which was backed by Ribbit, General catalyst and restive. ChatGPT can't move any of the money, right? So it's not like you don't have to worry like hey, it's going to steal my money or send it somewhere bad. It can't even see the full account numbers. It just can. It's a read only scope and financial data is excluded from any sort of training by default. It Intuit support is coming next, which means that your tax preparation, your credit decisions, all of that's on the roadmap. And So I think OpenAI is really trying to get into finance now. My take on this, number one, it's on the $200 a month tier, so basically no one's going to use this honestly. And number two is someone that would 100% use this. I am their target audience. At the beginning of this year I have an app I have which is called, used to be called truebill now, but was purchased by Rocket Mortgage, Rocket Money. So now it's called Rocket Money and basically it's the same thing where I have it linked to all of my credit cards and bank accounts and investment accounts. So it knows everything in there and kind of gives you a snapshot of like, hey, this is how much was on your credit cards, is how much in your bank, how much in your savings is what your net worth is, whatever. It's a very useful app and I've used it for many years. It has an export feature. So at the beginning of this year I exported all my transactions I made last year and I just gave that file to chatgpt and asked it questions. It was very useful. But there's no way I would pay $200 a month to be able to ask those questions when I could just click the export button and give that, you know, give that spreadsheet to ChatGPT. It takes like two seconds. So for $200 a month I think I will continue to push the export button whenever I want. And I guess if you want up to date information or it would save you pushing that button. But you know, I just kind of like to do a year end recap or maybe Monthly. You could do like a monthly every month, get a financial recap on where you were, like if you hit your goals, yada, yada. So assuming this isn't something that you need to do like every hour of every day, and pushing that update button is, you know, download button is too annoying, I wouldn't probably use this for that. But the other thing, I will say a caveat and to OpenAI's credit is a lot of times they'll throw something on the 200amonth tier to test it out and they bring it down to lower tiers. Actually that's kind of a bad strategy because, well, whatever, it's OpenAI strategy. They're the trillion dollar company, not me. But the problem with that is a lot of times I will see an update like this and if it was available to everyone, I would go set it up and try it. But because it isn't, I'm like, okay, well, I'll wait till maybe it gets down to everyone. By the time they announce it for everyone, I've moved on, forgotten about it. It's not in the news anymore. So that is a downside to the way that OpenAI sometimes does these rollouts, giving it to the $200 a month tier.
- [15:59]
B
First, if you'd like to get the top five AI news stories straight into your inbox every day, I have a newsletter called AI Chat Daily. It's on my website, AI chat daily.com I give a deeper dive into all of the stories on the show and I have a bunch of other content that I also publish over there. So if that's of interest to you, you can go check it out. There's a link in the description to check out the site and you can also subscribe there to get a daily AI newsletter. This will be the homepage for this show on the web and you can get all of the articles and content. It is aidaily.com if you want to
- [16:30]
A
go check it out.
- [16:31]
B
Okay, let's talk about what's going on with Greg brockman over at OpenAI. So OpenAI just told all of the staff that Greg Brockman is going to be taking over the entire product organization and he's going to keep his like, initial AI infrastructure role, but he's going to also be adding ChatGPT and Codex and the developer AI on top of it. All of those are basically being merged into a single core product team, which is, you know, some of the biggest stuff that OpenAI is in charge of and is shipping. And this is incredible because a lot of the, you know, original co founders of ChatGPT have left. Greg Brockman is still there and now he's getting more responsibility. So the original builder of Codex is thebalt Scoto and he is, you know, the person that's, you know, really in charge and kind of led the, had the lead on building Codex. He's getting promoted to lead that unified core product across consumer, enterprise and developers. Nick Turley, who scaled ChatGPT since it was launched in 2020 to more than 900 million weekly active users. He's going to move to a new role revamping enterprise products, but he's still going to be working on ChatGPT. Greg Brockman gave a memo over to this staff and the thing that I
- [17:40]
A
thought that was kind of stood out
- [17:41]
B
the most from that. He said, we're consolidating our product efforts to execute with maximum focus towards the agentic future to win across both consumer and enterprise. I think right now they know that they're racing towards an ipo. Anthropic is taking tons of enterprise share and this is an area they really have to push on. Putting Greg Brockman, he's the heavy hitter in charge of all these and consolidating them, I think is going to hopefully push OpenAI into kind of a new space and hopefully help them catch up to Anthropic. That feels like it is running away. And we just learned that Anthropic over, according to data from Ramp, has actually just passed OpenAI in in how much or what percentage of businesses are spending money on one of the two tools. Anthropic is now the number one choice according to the data. Okay, that's it for the show today. If you got something out of it, really, it would just help the show a lot if you could leave a comment, a rating or review. If you've listened to at least three episodes on Spotify, you can leave a rating. It's the about tab. You can drop some stars. I appreciate all of them. And on Apple, if you haven't left a review already, it would make me so happy and it helps the show get found by more awesome people like yourself. So if you haven't already, I'd appreciate that a ton. Make sure to go check out AI box.AI, my own startup, if you want to get access to over 80 of the top AI models in one place for only 899amonth. I use Claude to help me with writing, but it doesn't generate images or video or audio. I use 11 Labs for audio, Google, VO3 for video and ChatGPT for image,
- [19:06]
A
but I could do all of that
- [19:07]
B
on one platform at AI box AI for 8.99amonth. So I'll leave a link in the description. You can check that out. Thanks for tuning in and I'll catch you next time.