A long-term AGI roadmap puts agents first, then continual learning, then embodied AI
teortaxesTex · x · 2026-07-23
The roadmap on the slide: learning to learn, agents, then embodied intelligence
The image quotes a long-term AI roadmap that says the company’s goal should be AGI.
It argues that today’s systems already outperform humans when the problem is clearly specified with enough context, but they still lack two human-like traits:
- Continual learning: humans get up to speed after learning a company and role for a while, while AI does not.
- Real-world context: AI can only handle limited context and cannot yet replace employees in practice.
The roadmap described is:
- Learn to learn
- Agentic systems that expand what models can do
- Continual learning as the next bottleneck
- A possible singularity once models can keep learning and iterating on themselves
- Embodied intelligence after that, entering the real world to do housework and elder care
The speaker says each step builds on the previous one, and that the progression is more of a gradual slope than a sudden jump.
More from AGI Musings
- Moonshot says world models are outside its AGI mainline and not worth priority — teortaxesTex · 2026-07-23
- Moonshot figure’s surprise at Daya’s exit exposes a harder line on talent retention — teortaxesTex · 2026-07-23
- Scarcity in AI is now compute, domain expertise, and taste — rakyll · 2026-07-23
- Miles Brundage shares a clip arguing AI will push people into unprecedented collaboration — Miles_Brundage · 2026-07-23
- As models get more powerful, they become harder to contain — IgorKurganov · 2026-07-23
- A reply says Silicon Valley utilitarians often treat autonomy as nearly worthless — zetalyrae · 2026-07-23