AIs will achieve superhuman skill in novel domains like business or politics not by training on specific data from those fields, but by mastering the general skill of rapid learning and adaptation across millions of diverse, simulated RL environments. This skill then transfers to the real world.

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.

Jerry Tworek, a self-described "RL maximalist," found that scaling RL at OpenAI improved benchmarks but failed to solve real-world problems. The training data and evals were a closed loop, disconnected from the messy distribution of real user tasks, necessitating models that can learn at test time.

Standard benchmarks are too rigid. The future of model evaluation needs more open-ended, multi-agent scenarios like the "AI Village" project. Giving agents broad goals like "organize an event" reveals more about their "derpy" failure modes and real-world capabilities than constrained, benchmark-style tasks can capture.

Issues like 'saturation' and 'maxing' reveal a fundamental flaw: benchmarks test narrow, siloed abilities ('Task AGI'). They fail to measure an AI's capacity to combine skills to solve multi-step problems, which is the true bottleneck preventing real-world agentic performance and the next frontier of AI.

Beyond supervised fine-tuning (SFT) and human feedback (RLHF), reinforcement learning (RL) in simulated environments is the next evolution. These "playgrounds" teach models to handle messy, multi-step, real-world tasks where current models often fail catastrophically.

Traditional AI benchmarks are seen as increasingly incremental and less interesting. The new frontier for evaluating a model's true capability lies in applied, complex tasks that mimic real-world interaction, such as building in Minecraft (MC Bench) or managing a simulated business (VendingBench), which are more revealing of raw intelligence.

As reinforcement learning (RL) techniques mature, the core challenge shifts from the algorithm to the problem definition. The competitive moat for AI companies will be their ability to create high-fidelity environments and benchmarks that accurately represent complex, real-world tasks, effectively teaching the AI what matters.

Creating realistic training environments isn't blocked by technical complexity—you can simulate anything a computer can run. The real bottleneck is the financial and computational cost of the simulator. The key skill is strategically mocking parts of the system to make training economically viable.