Learning Machine is publicly confirmed only in outline — and Dwarkesh covers its exact thesis without ever naming it
Asked:
“Can you tell me anything about the company Learning machine founded my Yikang Shen on continual learning. Did Dwarkesh mention it?”
Eight sourced findings, gathered on 2 September 2026: three public traces of the company Learning Machine, and five instances of Dwarkesh Patel writing or asking about continual learning between February 2025 and August 2026. No source found mentions Yikang Shen or Learning Machine on Dwarkesh's site, podcast, or feeds.
Two tracks on one theme, never crossing
Dwarkesh's continual-learning coverage over time
Learning Machine's own homepage says it is building "the first generation of foundation models that can continuously learn, adapt, and improve" — the only first-party claim found.
learning-machine.ai
Yikang Shen's public profile lists him as CEO of Learning Machine and formerly member of technical staff at xAI; no page found explicitly names him founder, or states a founding date, funding, or team.
alphaxiv.org
Dwarkesh has escalated the continual-learning theme — from a question to Jeff Dean and Noam Shazeer, to calling it "a huge bottleneck", to a full essay of predictions — yet none of these five pieces mentions the startup.
dwarkesh.com
Absence of a mention is a search result, not proof: searches across Dwarkesh's site, transcripts, podcast indexes, YouTube and social results found no explicit reference — one could still exist outside indexed text.
dwarkesh.com
The evidence, row by row
| Track | Source | Published | Finding | Link |
Method: 8 evidence rows gathered by web research on 2026-09-02 — 3 public traces of Learning Machine (homepage, a startup listing, Yikang Shen's profile) and 5 Dwarkesh Patel posts or transcripts touching continual learning, each row a verbatim excerpt with its source URL. Dates are publication timestamps where available; two company pages carry none. Founder status, funding, investors, headquarters and launch dates were not verifiable and are not stated.