I did not find their description super accessible, but the Optech Podcast recap of something similar was helpful.

tl;dr is that BitVM advancements were bogged down by

but this new approach using garbled locks moves the vast majority of the transaction off chain and reduces staking requirements. I don't know if the approach Robin Linus describes on the Optech podcast is the same as what Alpen Labs is doing here, but it sounds similar. Linus notes that among "the tradeoff for this approach: it “requires a multi-terabyte offchain data setup”."