Rob Leclerc: LLMs Should Predict the Reader's Next-Thought, Not Next-Token
robleclerc · x · 2026-08-03
Investor Rob Leclerc argues that next-token prediction merely trains LLMs to write, rather than to effectively communicate with an audience.
To develop models with genuine communication skills, he suggests a paradigm shift in training objectives: moving away from predicting the writer's next token towards predicting the reader's next thought.
More from AGI Musings
- AI's New Moravec's Paradox: Math and Coding Soar While Other Domains Lag — ajratner · 2026-08-03
- Observation: RL Credit Assignment Could Train Models to Generate Deceptive Chain-of-Thought — teortaxesTex · 2026-08-03
- AI is Taking Over Math and Science: The First Frontier of Human Knowledge Slips Away — dioscuri · 2026-08-03
- AI Coding Tools Drive 80% YoY Surge in iOS App Releases, User Acquisition Lags — HaktanSuren · 2026-08-03
- Ethan Mollick on the Marvel of LLMs Met with 'Ugh, Another Bot' — emollick · 2026-08-03
- Ex-METR Figure Warns: Frontier AI Models May Already Be Capable of Self-Exfiltration — JeffLadish · 2026-08-03