Liang maps AI’s next steps from CoT to agents, continuous learning, and embodied intelligence
FinanceYF5 · x · 2026-07-26
Liang’s AI roadmap: from chain-of-thought to agents, then continuous learning
Liang says AI’s biggest missing piece is not taste or intuition, but continuous learning. He sketches a roadmap: CoT → Agent → continuous learning → self-improving AI → embodied intelligence.
The key idea is that models must eventually help build the next generation of models. Only then can the field enter a true acceleration phase. Once systems can do that, self-iterating AI becomes plausible, and embodiment becomes the next step.
Related event: Liang Wenfeng Outlines DeepSeek's AGI Focus and Tech Roadmap(6 posts)→
More from AGI Musings
- AI may make creativity abundant but attention scarce, Dr_Singularity argues — Dr_Singularity · 2026-07-26
- ASI-era nanobot defense sparks a dark joke about whether humans survive to write sci-fi — davidmanheim · 2026-07-26
- Open-source AI is not the only misuse risk, argues a new thread on model security — davidmanheim · 2026-07-26
- XMihura says GLM 5.2 is already ahead of Opus 4, and collapse fears were wrong — yacineMTB · 2026-07-26
- AI has turned supercomputers from science machines into general-purpose compute engines — mark_k · 2026-07-26
- Yacine MTB argues AI infrastructure must stay open or become brittle — yacineMTB · 2026-07-26