Large reasoning models almost certainly can think | Dark Hacker News