#249 - Fable 5 ban, SpaceX Cursor + IPO, OSS Aplenty

**Andrey Kurenkov** (0:11)
Hello, and welcome to the Last Week in AI podcast. We can hear us chat about what's going on with AI.
As usual in this episode, we will summarize and discuss some of last week's most interesting AI news. I am one of your regular hosts, Andrey Kurenkov. I studied AI in grad school and now work at the AI startup Astrocade.

**Jeremie Harris** (0:30)
And I'm your other regular co-host, Jeremie Harris, from Gladstone AI, AI national security, AI infrastructure, all those good things. And yeah, we're just talking about how this is a lighter episode than usual, not just because it's only five days instead of seven, since the last time we recorded, but also just like fewer papers, we're shooting the shit earlier or shooting the S as we say on the podcast, about why that might be.
I opened by saying, I mean, it makes me wonder how much things are getting, starting to get clamped down on a little bit, the labs in terms of how open they're being. But then Andrey just pointed out the sort of devastatingly clearly obvious reason why this is actually the case, which is conference season timing. So yeah, that's the... Not everything is a conspiracy, everyone. Not everything is a conspiracy.

**Andrey Kurenkov** (1:15)
Yeah, I'm a little bit out of a loop, but when there's a quiet period, it's a pretty safe call to say that it's in between major AI conferences. I mean, not to say that... I think we highlight really blockbuster papers often are from industry, Anthropic in particular, but also in some cases other labs. But we do cover quite a bit from academia as well. And there has been not as much what I've seen. Not to say that there's not been many papers, there's always a lot of papers that are interesting and worth discussing, just not ones that have gone semi-viral on like AI Twitter, I guess.
Yeah.

**Jeremie Harris** (1:57)
Actually, this is kind of since this shorter episode, maybe worth the digression here, but reading between the lines, the papers that I see Andrey picking and the papers at least speaking for myself that I ended up picking are generally, there's kind of a couple lenses that at least I find myself applying. The first is, will this paper materially change the macro picture? So papers around about scaling, you're seeing a result that actually, not just any random scaling paper, there's a million of those, but scaling papers that show trends that seem robust, that seem to apply to architectures where the hardware lottery is not going to be a massive problem. In other words, they can run pretty well on current hardware. That's a credible kind of paper where we will double down on those alignment results that seem, certainly anything that borders on like super intelligence, like super alignment grade alignment. Really interesting, same with interpretability. At least to me, that's kind of like the first order passes like, is this a paper that may change predictions around when AGIASI comes, recursive self-improvement, the risks and the outcomes of that process? That's kind of like the first order. And then there's like the kind of geopolitical, does this change US-China dynamics because maybe there's a breakthrough that spares us on memory, which means that one of the key bottlenecks that we have breaks, and now we can get help or vice versa with China. So I think that's a lot of like what comes to mind for me. And just kind of like economic transformation stuff too, but...

**Andrey Kurenkov** (3:21)
And I just like stuff that's interesting.
This information that seems novel and exciting, I guess, from a research perspective. So I guess together we have a nice rounded out set of things we like to focus on. We'd like to thank Box for being a sponsor. If you're a company trying to adopt AI, you're likely facing a common challenge. Most of the AI tools are great at public knowledge, but they don't actually know your business, your product roadmaps, your sales materials, your policies, all that stuff. And that's where Box comes in.
Box is building the intelligent content management platform for the AI era, serving as the secure essential context layer for Box's AI agents to access the unique institutional knowledge that makes a company run. The power of AI doesn't come from the model alone, it really comes from giving AI access to the right enterprise content. And that's where Box is perfect, it goes beyond file storage, it connects content to people, apps, and AI agents so teams can turn information into action.
With tools like BoxAgent, BoxExtract, BoxHubs, and more, organizations can accelerate knowledge work, pull intelligence from unstructured content, and automate workflows. So if you're thinking seriously about AI, think beyond the model. Your business lives in your content, and Box helps you bring that content securely into the AI era. Learn more at box.com/ai.