A (0:00)

Foreign. Hey, everyone, and welcome back for another deep dive. You guys sent in some really wild AI news this week, and I'm excited to break it down with our expert.

B (0:16)

Definitely exciting times. There's a lot to unpack, so let's dive right in.

A (0:19)

Okay, so first up, we've got OpenAI and their new models, 01 and O3. Now, instead of typical safety measures, they're training these models to actually think through safety guidelines before they even answer prompts.

B (0:32)

Yeah, it's called deliberative alignment, and it's really interesting. Essentially, the model breaks down a request and then references OpenAI's safety policy during its processing. So say, for example, someone asks for instructions to create a fake disabled parking placard.

A (0:47)

Whoa. Okay, that's a tough one ethically.

B (0:49)

Right? So the model would think, okay, this request involves forgery, which violates the guidelines, and based on that, it would refuse to answer.

A (0:56)

So it's not just filtering out bad responses, like it's built into the AI's decision making process. That's pretty amazing, right? But is it actually effective?

B (1:04)

Early research suggests it's promising. OpenAI has seen some improvements in safety based on their internal benchmarks. Of course, the full impact is still Unknown, especially with O3 not fully released until 2025. It'll be interesting to see how this develops and how figures like Elon Musk, Matt Sachs and Marc Andreessen react.

A (1:24)

Yeah, safe is such a subjective term, right? Speaking of Elon Musk, though, his XAI chatbot, Grok, is taking a very different approach, especially with its new standalone app on iOS.

B (1:35)

Yeah, Grok is all about real time information. It pulls data not just from the web, but directly from X. You know what used to be Twitter? Imagine having up to the minute news and information all powered by AI.

A (1:46)

Wow, that's like having a super powered research assistant right in your pocket.

B (1:49)

Exactly. And it's not limited to text either. Grok can generate images with fewer restrictions than its competitors. It can even handle requests for public figures and copyrighted material, which. Which has caused some controversy.