關於這一集
Before Smack, Andy Markoff planned wars in PowerPoint.
He was the operations officer for a special operations task force in northern Iraq during the fight for Mosul. Marine and coalition special operations teams were embedded with the Kurds, and later the Iraqis, and Andy’s job was to pull together what they were seeing on the front and figure out how to go after the ISIS network so the partner force could take the next piece of the city.
The tools were whiteboards and napkins. That got turned into a Gantt chart in PowerPoint. Every morning, SharePoint crashed trying to upload it. At the end of the night the plan went into what he calls the “special filing cabinet,” which is the trash can. For three months he slept one to two hours a night, “hopped up on energy drinks and rage.”
“There has to be a better way than this, but this is the way we have right now.”
It took him about seven years to get back to that problem.
About Andy
Andy knew he wanted to be a Marine before he got to Princeton. After graduation upon commissioning he grew frustrated by the inability to specialize. He moved from Rifle platoon to mortars, to aviation fires integration.
There, he learned that “most problems that are hard are integration problems. There’s very few problems that are worth solving that are single discipline problems.”
After he got out, he worked at Palantir, where they kept him off defense work on purpose because he was “too in the box” on the problem. Then it was off to Origin Materials, where he ran operations from Tahoe and guided in the mountains on weekends. Then Pallas Performance, a sports performance company he started with his wife.
About Smack
In fall 2023, Andy’s friend Clint was leaving the Marine Corps, and the two of them decided to go after the very problem that had been troubling them for years. Through the Palantir network they met Dan Gould, the former CTO of Tinder.
Here’s how Andy frames the problem: national leadership sets strategic goals. Units carry out tactical actions. Everyone talks about one or the other. The layer in between requires arranging all those tactical actions in space and time and tying them to strategic effects you can measure. But that layer still happens on napkins and whiteboards and in PowerPoint.
“We can’t campaign. There are no tools that exist that enable campaigning.”
So Smack builds small, domain-specific models that run on-prem in compute-constrained environments.
The harder part is the training data. LLMs learn from text. Andy’s point is that very little of what matters here is written down. How platforms, effectors, and signatures interact at scale, given a mission and a set of objectives, lives in the heads of people who have done it in combat and taught it. So Smack has to build a dataset that doesn’t exist yet, combining physics models with knowledge pulled out of those people’s heads. Andy calls the output trajectories. Some are fully synthetic. Some are hybrid, with real post-mission data filling part of the schema and synthetic data filling the variables nobody measured.
For more on Smack, visit: https://smacktechnologies.com/
Follow Andy: https://www.linkedin.com/in/andy-markoff-0b096611a/
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.valleycrossers.com
