AI Builds Itself as the Spending Boom Accelerates

SPEAKER_1 (0:02)

Bloomberg Audio Studios, Podcasts, Radio, News. Bloomberg Tech is live from the heart of Silicon Valley with Ed Ludlow in San Francisco.

Ed Ludlow (0:23)

This is Bloomberg Tech, coming up as the global debate around AI heats up. Anthropic finds its Claude Chatbots leading 26% of its AI research and development. Plus SoftBank's closing out the week with nearly $21 billion in potential fresh borrowing as it builds out its AI financing capacity. And Crusoe CEO Chase Lochmiller joins us on the race to build the infrastructure powering AI. The company closes nearly $4 billion in fresh funding. We're going to end the week much as we started it talking about Anthropic. Anthropic says it's Claude Chatbot drives more than a quarter of the company's research and development work for AI and collaborates with staffers for about 90% of their work. It's the clearest indication yet that AI can help speed up development of future models. Bloomberg's Shireen Ghaffari has the data. And Shireen, just start by explaining the data that Anthropic disclosed.

Shireen Ghaffari (1:17)

That's right. So Anthropic is basically measuring how much its own researchers, its own R&D department is using AI in its work to create new AI models. And what they found is that right now, Claude is leading, Claude meaning Anthropic's AI, 26% of that AI R&D work.

Now, that's not the same as AI doing everything by itself, but that is significant. It's one of the few kind of numbers that we have quantifying what many labs like Anthropic Open AI are now moving toward, which is the AI essentially feeding the AI to keep improving itself.

Ed Ludlow (2:00)

We're at the end of a seven-day week, even though it's Friday, because on Saturday, seven days ago, Dario Amode published his essay.

And one of the things he highlighted at the top for the rationale for doing it now is recursive self-improvement, the idea that AI is making the training of future models better. But do you have an answer to that, Shereen, why the data that was published yesterday came out when it did?

Shireen Ghaffari (2:26)

You know, I think we don't know if there is any one specific reason, but I will say the backdrop of all of this is while that there is obviously huge business opportunity in reaching AI that can improve itself with RSI, there is also huge concern, anxiety, risk. And we're seeing AI lab employees at Anthropic, at OpenAI starting to raise concerns publicly about this. We saw the viral note of Jacob Cox in a former Anthropic and employee researcher saying, he doesn't think any of the labs are handling this responsibly right now.

And I think that there's a real call for these companies to start quantifying and potentially slowing down by Dario Amadei's own essays recognition. The pace of this, even if that means losing some business momentum.

Ed Ludlow (3:18)

Author of the Bloomberg Q&A newsletter to which you subscribe, Bloomberg Sharim Khafari. Thank you very much indeed. As AI starts helping build the next generation of AI, warnings about losing control are growing louder. But critics say the focus on existential risk is overshadowing problems that are already here. It's the subject of Bloomberg, Rachel Metz and Davy Alba, latest reporting and frankly, great body of work trying to cut through the noise of the week. Davy, I'm going to start with you. Just summarize what our story outlines and take us to the origin, which is the hugging face incident.

Davy Alba (3:56)

Yeah, absolutely. So, you know, we did a deep dive into sort of the hugging face incident as an example of how a lot of the rhetoric that's been going around right now is talking about how a failure in safety and in the sandbox not being properly cordoned off, that there wasn't warnings. There were warnings that the human employees themselves did not take seriously. That is being framed as, look, we can't contain this AI, it's so dangerous, it's going to escape, and that being an inevitability.

And that just sort of works to sidestep the responsibility that these companies have to avoid safety failures.

Ed Ludlow (4:46)

Hey, Rachel, you know what's weird about AI?

What's weird? Humans, peeps, designed the experiment and designed the safeguards. That's the takeaway I got reading your piece. Remind everyone that AI came from humans. Run with it.

Rachel Metz (5:05)

Yes.

Yeah, I think that's something that I've been thinking about for a while, since we heard about the hugging face in Invert and Hack, and talking to people about it. I mean, this comes up quite a lot, but humans were very much involved with this. I mean, humans set goals for AI. Humans set up the situation here in the sandbox, where the AI is working on things, and humans are training. I mean, there are humans up and down. That humans also noticed some situations where they could have stepped in during evaluations when the AI wasn't doing quite what it should have been doing, and did not. And OpenAI has acknowledged all of these things, and to its credit has said quite a lot about this incident, and also what it's doing to change things over there. But I think that that came up a lot with this story, especially as we're talking to people ongoing, that people have a lot to do with this situation. It's not as simple as saying, oh, well, AI went and did something.