xAI Grabs Cursor and Sakana Goes Deep

**Beth Lyons** (0:00)
Hey everybody, good morning. Good morning, good morning. Happy Tuesday. Happy AI Tuesday. This is The Daily AI Show. I am Beth Lyons. With me in the studio today is Andy Halliday and Anne Murphy. We expect Brian to come in the door at some point. It is Tuesday, June 16th, and it's an episode in the late mid 700s.
It ends, I believe, in 7 Are we in 7.57 maybe? 7.52. We'll let you know. Brian will let you know what episode we're doing. Sorry, I did not check that before I came on. Welcome everybody in the chat. Jeff, Brian, Jen, so good to see you, and all the numbers of people who don't join and say hi. We're super glad to have you as well.
Still no Fable today, but still lots of conversations around it.
We usually start. Andy, do you have something specific? What's alive for you today? Then we'll go to Anne.

**Andy Halliday** (1:13)
Sakana Marlin.

**Beth Lyons** (1:15)
Sakana Marlin, our buddies over in Japan. Absolutely.

**Andy Halliday** (1:18)
Yes. Sakana has been doing a lot of deep research, you know, and advancing evolutionary model development and has recently launched a recursive self-improvement lab as an approach to add to their evolutionary models. But they've now delivered, they previously delivered a non-commercial product for science, which was an AI scientist.
Now they're delivering their first commercial product, which is aimed at high-level strategic decision making.

**Beth Lyons** (1:58)
Yeah.

**Andy Halliday** (1:58)
And it is focused on intense strategic research that a chief strategy officer at a major company might undertake, that they would typically do with a team of several people over several weeks in order to come to the CEO and the board of directors with a decision framework that the team could use then to make key decisions that impact the future of the company and its survival and thrival.

**Beth Lyons** (2:29)
And I love when they announced it, they're like, this isn't deep research, right? It's not like everybody else's mirror deep research. This is strategic research, which is an important distinction.

**Andy Halliday** (2:45)
Which alludes to the breadth of research that it does, and the analysis and reasoning that is applied to that collected research. But the thing that's made this possible is long horizon autonomy by the agents that are involved in this process. So it's able to stay coherent over an eight hour period. And that's, agents can do sort of work using a big project folder, and they can continue working based on referring back to a plan that already exists. But to do this kind of intensive research and synthesize it into a report using thousands and thousands of resources and staying on track and figuring out what's meaningful and what's not, and delivering that in a sort of crystallized way, that's a new thing. And I'm not surprised that Sakana has come forward with it. So it's a simple tool to use for the strategic team. You just say what the research theme is.
Hey, we have a new competitive entry in this product space. We're losing market share in this area. So what's the answer?
How do we offset the gains that are being made by that product? And it's now going to go out and do every kind of research related to that sort of thing.
But before it does that, it's going to refine the objectives with dialogue that it conducts with the key strategic players in the company. Once you've set the direction, no human intervention is needed after that. It goes out and autonomously repeats hypotheses, and then information gathering, and then verification or deferral of that hypothesis. And it goes beyond summarization when it comes to a result. It organizes all the causal relationships in the complex business environment, structures those things together as strategic options, and then delivers that to management so that they can consider that. They did a closed beta with 300 people who are serious players in this space. And that started in April of this year, and so it was just a few months long, because of course Sakana built something that can do this in eight hours. So they got a lot of action from those 300 people.
It goes, it did that test, and then they were from diverse industries, financial institutions, consulting firms who often do strategic analysis for their client companies, think tanks, and then they announced it. Yesterday, I wanted to just read in the conclusion in this generous and honorific way that Japanese people phrase things, but it's not Japanese people behind this company, remember, these are two Google, ex-Google people, but who loved Japan and moved to Japan to do this. But a lot of their output is done in Japanese, so I'm sure that these two Googlers love Japan a lot because they're quite fluent in Japanese. They're not there for any other reason than that. I'll be interested in discovering that.