GOOGL

<p>DeepMind scientist Jon Richens presented a paper at ICML 2025, proposing the view that agents are essentially world models, arguing that achieving AGI requires learning predictive models of the environment. Shangmin Guo from the University of Edinburgh supports this conclusion and points out that strategies can be unified with world models into a single LLM. The research also aligns with Ilya's 2023 perspective, emphasizing the importance of world models in AGI. Although there are model-free agents, whether they have learned implicit world models remains to be explored</p>

- DeepMind's recent paper presented at ICML 2025 argues that achieving human-level intelligence (AGI) requires a world model, revealing that agents inherently learn predictive models of their environments.  
- The study concludes that any agent capable of generalizing across multi-step goal-oriented tasks must have learned an accurate world model, which is essential for improving performance on complex tasks.  
- This research aligns with previous findings and suggests that there is no shortcut to achieving AGI without learning a world model, emphasizing its critical role in intelligent behavior.

DeepMind reveals astonishing answer: Agents are world models! Coinciding with Ilya's prediction from 2 years ago