Skeptics warn that sigmoid flattening limits extreme AI research acceleration
In a post on X, Will Brown argued that the staying power of DeepSeek-V3's architecture should force a rethink for people expecting unbounded algorithmic gains from recursive self-improvement. In follow-up replies on X, Brown said ideas like a "1000x R&D speedup" are too loosely defined, and argued that architectures, optimizers, infrastructure, kernels, chips and algorithms all hit sigmoid-shaped limits rather than rising forever (reply, reply).
Combined views
8.7K
7 posts, first seen 3h ago
Skeptics warn that sigmoid flattening limits extreme AI research acceleration
In a post on X, Will Brown argued that the staying power of DeepSeek-V3's architecture should force a rethink for people expecting unbounded algorithmic gains from recursive self-improvement. In follow-up replies on X, Brown said ideas like a "1000x R&D speedup" are too loosely defined, and argued that architectures, optimizers, infrastructure, kernels, chips and algorithms all hit sigmoid-shaped limits rather than rising forever (reply, reply).
the longevity of the DSV3 architecture should be an update to anyone whose world model hinges on unbounded algorithmic progress via recursive self-improvement
