Dev Slams 'Swarm' Narrative: Trillion-Dollar Valuations Built on a Sampler With No Brakes
gerardsans · x · 2026-09-23
Gerard Sans sums up a debate over how to name agent systems and their failure modes: 1) The reason to care about what we call these systems is to understand how they fail; 2) These 'processes' aren't ordinary software—they're samplers that fail in predictable ways (often silently), while the harness driving them has no brakes—brakes were removed to sell the story of RSI takeoff, autonomous labor, and trillion-dollar valuations; 3) Putting brakes on the samplers would pop the AI bubble; and another function of the 'swarm' narrative is that when samplers fail silently at some rate, the bill can be passed on to users.
More from AGI Musings
- gLM2 genomic language model designs new-to-nature biosynthetic enzymes — anshulkundaje · 2026-09-23
- CAIS: AI can now automate 20.8% of remote labor projects, up from 2.5% a year ago — ZeroStateReflex · 2026-09-23
- Peter Diamandis: AI companies must own their releases—speed and accountability can coexist — PeterDiamandis · 2026-09-23
- Intelligence may become abundant before power plants and fabs can keep up, argues thread — r0ck3t23 · 2026-09-23
- Guardian essay: a multilingual academic's complicated love affair with AI — nordicinst · 2026-09-23
- Monitoring AI agents' bookings takes as much bandwidth as doing it yourself — nabeelqu · 2026-09-23