Intelligence lives in the geometry of information, not the weights, argues thread
yunta_tsai · x · 2026-10-04
The author argues models can carry similar bytes in their weights yet differ in perceived intelligence—the difference lies in how information is organized geometrically, which reward processes can improve or disrupt. Extending the analogy to humans: intelligence is a reversible process that regresses when beliefs go untested and sharpens when grounded in reality-based feedback.
More from AGI Musings
- 6M US car crashes a year: why FSD could make accidents optional once insurers price it in — GabGarrett · 2026-10-05
- Jensen Huang Calls Musk and Hinton's AI Doom Talk 'Irresponsible' and Unscientific — ylecun · 2026-10-05
- US AI-Linked Jobs Surpass 750K Since 2023, With ~$180K Median Salary — DavidSacks · 2026-10-05
- 55% of American adults take zero flights a year — a stat that stuns AI insiders — SuB8u · 2026-10-05
- AI safety researcher: LLMs aren't conscious, but the viral article's arguments are bad — davidmanheim · 2026-10-05
- Consciousness lives in the pattern, not the substrate: an introspection argument — RileyRalmuto · 2026-10-05