AI models show impressive performance on evaluation benchmarks but underwhelm in real-world applications. This gap exists because researchers, focused on evals, create reinforcement learning (RL) environments that mirror test tasks. This leads to narrow intelligence that doesn't generalize, a form of human-driven reward hacking.
The popular concept of AGI as a static, all-knowing entity is flawed. A more realistic and powerful model is one analogous to a 'super intelligent 15-year-old'—a system with a foundational capacity for rapid, continual learning. Deployment would involve this AI learning on the job, not arriving with complete knowledge.
The central challenge for current AI is not merely sample efficiency but a more profound failure to generalize. Models generalize 'dramatically worse than people,' which is the root cause of their brittleness, inability to learn from nuanced instruction, and unreliability compared to human intelligence. Solving this is the key to the next paradigm.
AI models excel at specific tasks (like evals) because they are trained exhaustively on narrow datasets, akin to a student practicing 10,000 hours for a coding competition. While they become experts in that domain, they fail to develop the broader judgment and generalization skills needed for real-world success.
The fundamental challenge of creating safe AGI is not about specific failure modes but about grappling with the immense power such a system will wield. The difficulty in truly imagining and 'feeling' this future power is a major obstacle for researchers and the public, hindering proactive safety measures. The core problem is simply 'the power.'
The era of guaranteed progress by simply scaling up compute and data for pre-training is ending. With massive compute now available, the bottleneck is no longer resources but fundamental ideas. The AI field is re-entering a period where novel research, not just scaling existing recipes, will drive the next breakthroughs.
Emotions are not superfluous but are a critical, hardcoded value function shaped by evolution. The example of a patient losing emotional capacity and becoming unable to make decisions highlights this. This suggests our 'gut feelings' are a robust system for guiding actions, a mechanism current AI lacks.
It's a profound mystery how evolution encoded high-level desires like seeking social approval. Unlike simple instincts linked to sensory input (e.g., smell), these social goals require complex brain processing to even define. The mechanism by which our genome instills a preference for such abstract concepts is unknown and represents a major gap in our understanding.