AI That's Smarter Than Humans? Not Anytime Soon

Watch AI That's Smarter Than Humans? Not Anytime Soon - Bloomberg on Dailymotion

Transcript

00:00First, OpenAI announced it was pivoting to porn, then it treated information it had scraped from

00:06the internet as huge news. Those two things might not seem related, but they do point to how the

00:11company's ambitious goals of benefiting humanity are starting to look ever more hazy. Let me tell

00:17you about this latest blunder. One of OpenAI's leading scientists recently posted this on X.

00:24A lot of people, understandably, thought that was pretty misleading, and he eventually deleted it.

00:30Sam Altman has long argued that large language models, or the technology that underpins chat

00:36GPT, which scrapes information from the internet and synthesizes it into clever answers, is the

00:42path towards AGI, or Artificial General Intelligence, this theoretical threshold where computers are

00:49smarter than humans. But that's not what happened with these unsolved math problems known as Erdos

00:55problems. The researcher who literally runs the website that keeps track of all of those said as

01:01much. GPT-5 had just found the solution somewhere else on the internet, but the researcher just hadn't

01:07updated them on his website. Even rival AI labs started piling on. This is all a stark reminder

01:13that large language models still aren't very good at reasoning, but they're good at pretending

01:18to reason. They're ultimately glorified pattern matching tools, which can make money from sex

01:25bots or AI slop. Don't get me wrong, chat GPT is still quite useful. It's used by 10% of the world's

01:32population, for many of them, as an alternative to Google search. That can be lucrative, especially

01:37when it involves giving users access to porn. But that's not the kind of ambition that led to

01:43open AI achieving a $500 billion valuation. Don't expect that to stop Silicon Valley from telling

01:50us that superintelligence is just around the corner, because the AI boom is going to continue

01:55to run just as much on hype as it does on genuine breakthroughs.