AGI: The Coming AGI Wars: Players and Positioning

ONE SENTENCE SUMMARY

Analyzes three AGI pathways—LLMs with RLHF, Yann LeCun’s RL, Carl Friston’s active inference—evaluating players, funding, technologies, futures, and trajectories ahead.

MAIN POINTS

1. LLMs offer capabilities but lack general intelligence; they rely on conditional probabilities and risk hallucinations.
2. Knowledge graphs and world models may overcome LLM intrinsic reasoning limits.
3. Massive funding fuels OpenAI, Google, and Meta; dozens of mid-sized players compete.
4. RAG and RLHF augment LLMs, but they don’t provide AGI on their own.
5. LeCun champions RL and world models; Meta’s Fair lab advances not solely LLM-centric.
6. Friston’s active inference minimizes free energy over time via variational inference.
7. CORTECONs (corticons) bridge signal-based AI with ontologies for grounding.
8. Expect 2-3 years before notable AGI progress becomes visible.
9. Altman’s visions: AGI as helpful assistant or autonomous senior colleague.
10. Different AGIs will feel distinct, changing user interaction dynamics.

TAKEAWAYS

1. AGI progress hinges on grounding, ontology, and world models, not scale alone.
2. Three camps compete: LLM-based, RL-driven world models, and active inference.
3. Huge investments shape leaders, but agile disruptors can emerge from smaller firms.
4. Foundational knowledge in stat mech and variational inference supports AGI work.
5. AGI flavors influence user experience and decision-making in real applications.