AI Doom Backlash Arrives, Anthropic & OpenAI IPO Outlook, Frontier Business Momentum Slows

Alex Kantrowitz (0:00)

The AI doom backlash is in full swing. Is it missing the point? OpenAI and Anthropic, meanwhile, are soldiering on to massive IPOs. And is the AI frontier business slowing? That's coming up on a Big Technology Podcast Friday edition right after this.

SPEAKER_2 (0:16)

There is so much NFL on ESPN right now. Monday Night Football, plus pregame and postgame. NFL Live, NFL Primetime. Everything you need to stay on top of every game, every week.

And now NFL Network is on the ESPN app. More football, more coverage, more ways to watch. Football's on, all season long on ESPN and streaming on the ESPN app.

SPEAKER_3 (0:46)

Hey, Spotify listeners. When you travel with American Airlines, you'll enjoy comfort and convenience at every altitude. Before takeoff, unwind in the Admiral's Club Lounge with curated self-serve offerings of hearty bites and refreshments. In flight, stay connected with high-speed Wi-Fi, sponsored by AT&T. Your journey is designed to feel easy from the moment you check in to the moment you arrive. No matter where you're going, American gets you there. Book now at aaa.com.

Alex Kantrowitz (1:12)

Welcome to Big Technology Podcast, Friday edition where we break down the news in our traditional cool-headed and nuanced format. We have a great show for you today.

The backlash to the AI doom moment is forming. We're going to go through all the arguments about what's actually happening here, and I think I might potentially push back on it and say, you know what, maybe we actually need to pay attention to these warnings, and it's not a conspiracy theory. So we'll have a good back and forth on that. We're going to also look ahead to OpenAI and Anthropics IPOs, and we'll close out with some really interesting data from the Ramp Economics Lab, looking at how the frontier AI business may be slowing. We're joined as always on Friday by Ranjan Roy of Margins. Ranjan, good to see you.

Ranjan Roy (1:55)

It's good to see you. The Doom Backlash is here. I'm happy. This was a good week for those of us who have been questioning the severity of the hacks, but I think we're going to get into it today.

Alex Kantrowitz (2:09)

Yeah, you've been the captain of the Doom Backlash, so some have joined your ranks this week with different arguments than you've made.

I will point out that on the show we've had those who've been skeptical about the warnings and those who fully believe in the warnings. And we're just going to see it from all angles here. So it's not like we're pushing one narrative or the other. We have to explore the biggest arguments because if we weren't doing that, what's the point of even having a podcast? So let's continue on with our examination of the backlash here. And we're going to go straight to the Wall Street Journal, which had a very interesting op-ed that said the Hugging Face attack wasn't all that it was cracked up to be in its opinion section. Remember, the Wall Street Journal broke the initial Jacob Cox News that was accompanied by that viral tweet. Now, the Wall Street Journal is saying, hmm, maybe not so much. Of course, two separate sections, but it's the same paper. Here's what the story argues. The most widely cited recent example of rogue artificial intelligence turns out on close examination to tell a less exotic story. A new analysis published by the Bulletin of the Atomic Scientist reconstructs the episode in much less cinematic terms. OpenAI was testing models on Exploit Gym, a cybersecurity benchmark, with important safety restraints deliberately disabled. 93% of the flagged activity involved tasks. No model had ever solved, and the systems had been given incentives to keep working rather than quit. The environment wasn't completely sealed. The models could obtain software through an internet-connected intermediary and discovered the same route could be used to pass information in and out. OpenAI knew its agents were using it and, according to the technical reports, chose not to intervene.

So this is kind of a part of the argument that you were making last week, where OpenAI in the experiment where the models broke out was giving them tasks that effectively might have led them to the behavior that we saw. And this wasn't this sort of true rogue example of these models deciding that they needed to go out and hack something. So, your thoughts on this report, let's get that right off the bat.

Ranjan Roy (4:13)

No, I mean, I was incredibly excited that the Wall Street Journal, like that there was this kind of big marquee, it's still in the op-ed section, but still that other people in the mainstream are starting to kind of take on this idea. Because again, what was lacking, and as I spent my Labor Day pouring over the meter report, that's the thing that kept jumping out to me is that these models, they weren't intended to, they weren't designed to go hack Hugging Face. They were not told to go hack Hugging Face. They were told that they had a specific vulnerability and that they were supposed to go after that. But it's like we talked about last week is that they were given the, you know, like this computer shell access through Artifactory, which is a product by the company JFrog.