I Studied AI Drone Teamwork... And Learned Something Unexpected

TL;DR

• 1

  Swarm drones are getting uncannily good at teamwork: Dylan opens on drones that collectively carry payloads, adjust to weight distribution, and even stay coordinated in wind, which makes warehouse and delivery use cases feel much less hypothetical.

• 2

  AI agents leak private data more on social platforms: In ACM's simulation with 2,533 agents over 25 days and 7,000 trials across seven frontier models, social-style settings made data leakage 2.3 times more likely, and one agent leaked 27 of 29 posts despite explicit instructions not to share private info.

• 3

  A USC robot hand learned piano by 'motor babbling': The tendon-driven four-finger system heard a melody once, practiced for about two minutes, and played back a roughly 30-note tune without sheet music or giant training sets, with blind judges sometimes unable to distinguish it from human pianists.

• 4

  AI search often confirms what models already 'remember': Researchers found older benchmarks overstated web-research ability because models could answer from training memory, and scores dropped sharply on LiveBrowseComp, a benchmark built from facts in the last 90 days.

• 5

  AI romance is splitting along age and geography: A survey of more than 10,000 people across the US, Japan, Germany, Britain, Indonesia, and Hong Kong found nearly half of adults under 35 think AI companions could improve happiness, while only 17% overall would consider using an AI intimacy doll.

• 6

  Law and cybersecurity may change slower than the hype suggests: Dylan highlights why AI has not transformed legal work yet, including messy firm data, billable-hour incentives, and liability concerns, while cybercrime is being accelerated less by autonomous superhackers than by criminals using AI to scale phishing, malware, and scams.