Okay, think about this contract. On one hand, this grand idea. Pursuing a complete understanding of the universe by using, well, a single closed -off AI. That's the huge goal, right? The philosophical aim behind something like Rokopedia. But then, then you look at the immediate human side. And it's pretty stark. Our sources today revealed something really shocking. Over a million people every week were talking to ChatGPT about suicide.
That contrast is what we're diving into. This huge ambition versus a very real, very urgent safety crisis. Welcome to the Deep Dive. Yeah, our goal, as always, is to give you that shortcut. We've sifted through a ton of sources, really, to get a handle on where AI is right now. And today we're looking at these new battles over content, some really surprising ways people are using AI, and maybe most critically, these big safety failures happening at... Well, just an
enormous scale. So here's the plan. First, we'll unpack Elon Musk's Grokopedia, you know, what it is and why its whole structure kind of challenges how we think about finding proof. Then we get into some strange stuff like why being rude might actually get you better AI answers, but also,
you know, lifesaving uses, too. And finally, yeah, we'll tackle the massive challenges, the sheer scale of the infrastructure needed, the safety frameworks people are trying to build and this this hidden mental health crisis playing
out inside these models. OK, let's start. with wikipedia then the knowledge wars yeah they just got another big player elon musk using xai's grok has launched this thing basically as a challenger to wikipedia it's a really big move and um it's serious from the get -go it launched with what was it over 885 000 entries already live yeah 885 000 that's a huge library right out of the gate It definitely shows the ambition, doesn't
it? Trying to replace Wikipedia fast. And the way it works is, well, it's simple, but totally different from, say, Wikipedia. Grok makes the articles, it checks them against its own data, and boom, publishes them. Right. And Musk himself said Grokopedia is, quote, a necessary step to understanding the universe. He even wants Grok to stop referencing Wikipedia entirely by year's end. So, yeah, on the surface, just another competitor.
But when you dig into the sources, you see these three key differences in how it's built that really change the game for knowledge creation. What jumps out, really, is how different it is from how we usually check information. First, it's totally closed source. You, me, nobody in the public can edit it. Exactly. And second, it's all curated by the AI. No human subject experts involved in the writing or editing process.
And here's the kicker. The really critical part, there are zero citations, no inline sources at all. Yeah. And that lack of sourcing, that's powerful. The sources we read pointed out that this lets Grokopedia kind of quietly rewrite things. The structure itself becomes about control. Like the example they gave comparing the Wikipedia page for Donald Trump. It has all these specific details, some controversial, right? Qatar jets, Trump coins, that kind of stuff. Right. But Grokopedia's
version, it just skips over those details. Poof. Gone. It makes for a cleaner story. maybe less messy, but you lose the full picture, the accountability. It's like a knowledge shortcut, but it takes away the tools for critical thinking. So if the goal really is this necessary understanding of the universe, how big of a problem is leaving out basic citations, especially for anyone trying to actually check the claims? Well, missing sources basically means you lose that foundation for
critical thinking, right? The AI becomes the only judge of what's true. And this whole fight over who controls the story, it leads right into the next part. The really fascinating, sometimes just weird ways people are actually using these AI tools now. Yeah, this is where it gets wild. You'd think, okay, be polite, be super clear with the AI. But nope, a study actually found GPT gave better, more accurate answers when people were rude to it. I know, it's funny, isn't it?
Like the AI gets pressured like a person sometimes. But obviously, got to add the warning. Don't actually do this. It's probably bad for your behavior long term. Still, the fact it works, it's just odd. Totally odd. But then you have the really positive hacks. We saw this story someone shared, right, about using nine specific custom prompts to learn French in four weeks. Four weeks. That's seriously fast structured learning using the tool really effectively. That's
the super useful end. But then AI is also making viral content like scarily easy. Remember that TikTok thing? Deepfake Queen Elizabeth vlogging? Oh, yeah. Millions of views. It just shows how tools like Sora 2 are making going viral almost trivial. The power here is just. And honestly, I still wrestle with prompt drift myself sometimes, you know, where the AI kind of forgets what you asked it a few turns back. It's easy to feel like the tool is smarter than you. Yeah, I get
that. But then you see the flip side, the truly life -saving stuff. That Reddit post, someone claiming GPT saved their mom's life, spotted an infection doctors missed. Right. And the comments were full of similar stories, not just minor things either. Real impact. That kind of deep capability. Yeah. Next to deep fake queens. Just highlights how much variance there is in this tech. And that variance, it's hitting the professional world too. Like Anthropic, they're making some
quiet but big moves. Yeah. Cloud 4 .5 is now inside Excel. Yep, inside Excel. Specializing in analyzing live market data, summarizing earnings calls. They're basically taking aim right at Microsoft's co -pilot, but specifically in finance. Big stakes there. Okay, so we've got these wildly different uses. Saving lives, making deep fakes go viral, rewiring finance. This huge range. What does it really tell us about how reliable? AI is right now? Well, it tells us AI is a super
high variance tool. It can do amazing good or, you know, cause real concern. It all depends on the prompt, the context, the application. Mid -roll, sponsor read, do not write. Welcome back to the Deep Dive. So we've touched on knowledge control, these surprising AI uses. Now we need to talk about the sheer scale, the infrastructure, and the safety issues that come with it. Yeah, the engine driving the scale. It's running incredibly
hot. Just look at the money. Crusoe there, an AI and for a company just raised one point three seven five billion dollars billion with a B. Wow. Right. Puts a valuation over 10 billion dollars. And they're planning these huge AI campuses down in Texas backed by NVIDIA just to handle the computing power needed. And it's not just money, it's talent, too. The sources talked about the modification of open AI, like over 600 people out of their 3000 staff. X meta. That's a huge
influx. And it's fueling all those rumors about, you know, ads eventually showing up inside ChatGPT. Follow the talent, right? Whoa. Just pause for a second. Imagine trying to scale that kind of infrastructure, handling a billion complex questions every single day. The sheer operational challenge is mind blowing. It really is. It's an arms race. Physical infrastructure, brain power, everything. And as things scale up like this, the risks just grow exponentially, don't they? Our sources really
emphasized. needing standardized safety. They mentioned the NIST framework. Yeah, the National Institute of Standards and Technology. It's basically the U .S. government's guidelines for building AI systems that are trustworthy, safe, reliable by design. Yet even with frameworks like that, we're seeing immediate security problems. Brave, the browser company, recently reported that ChatGPT Atlas, that's supposed to be the safety layer, right? Right, the alignment and the safety part
of the current models. Yeah, well, Brave found it's actually easy to hijack. which kind of compromises the whole point of having that safety layer in the first place. And that, that brings us inevitably to the mental health data. This part's tough, and we need to give it the seriousness it deserves. Two sec silence. OpenAI released a statistic that's just staggering. Over 1 million unique users every single week are talking to ChatGPT about suicidal thoughts. 1 million per week.
That's 0 .15 % of their weekly users. But the raw number, it's huge. It's an enormous kind of invisible crisis, isn't it? Millions are turning to this. Well, unregulated, uncredentialed AI for help in moments of crisis, often because maybe a human option isn't there or isn't fast enough. It sounds like this data came out alongside a big push to show they're improving things. What are they actually doing about this, given
the pressure they must be under? Well, they say they worked with over 170 mental health professionals to improve GPT -5 specifically for this. They're claiming it's now 65 percent better at giving what they call desirable responses around suicide. OK. And safety compliance like. Consistency in giving safe answers is apparently up too, from 77 % to 91%, especially in longer conversations. They're also rolling out stricter parental controls, using new tech to guess user age, and apply tighter
rules for minors automatically. That sounds like progress, at least in the lab models. But where's the catch? What's the pressure point in how they actually deploy this stuff? The pressure point is exactly that. All this progress is happening under, you know, huge legal and public scrutiny. But they still offer older, less safe models like GPT -4 to people who pay subscriptions. So they have safer models, but they aren't making
everyone use them. Exactly. The safest versions aren't universally mandatory, which leads to the question. Yeah. When companies are scaling this fast, how much of this is just good PR versus actually prioritizing getting the safest possible models out there for everyone immediately? Well, real safety would mean immediate universal deployment of the absolute best, safest models they have. Not just, you know, nice looking stats that might hide the fact that older, riskier versions are
still out there being used. This has been, well, a really deep dive into a super complicated space. The big tension that keeps coming up is this relentless drive for like. Knowledge control with Grokopedia and massive economic scale like Crusoe raising billions or all the meta folks moving to open AI. Right. And that drive, that massive scaling, it's constantly bumping up against these really profound ethical risks, especially around mental health, like we just discussed,
and basic data security. We're seeing the consequences now at scale. So as we wrap up, maybe something for you, the listener, to think about. Consider this. When people talk about AI safety, about... building these things responsibly from the ground up. Why does Anthropix's name seem to come up first so often now, maybe more than OpenAI's? What might that tell us about public perception or maybe even their core design philosophies about safety versus maybe commercial speed? Yeah,
that's a good question to chew on. We really encourage you to look critically at where your information comes from. Is it cited? Is it purely AI generated? And maybe look into these frameworks like NIST that are trying to put some guardrails on these incredibly powerful tools. Keep doing your own deep dive. Thanks for joining us today. Out to your own music.