$AILABS - AI labs are making a major research bet that training on millions of verifiable tasks across diverse RL environments will create AGI-level problem-solving agents. This represents the core investment thesis driving current AI development.
$INFRA - Inference compute represents 30-50% of lab compute budgets and will become increasingly valuable as continual learning and test-time training ('dreaming') emerge as a fourth axis of scaling alongside pre-training, RL, and inference time compute.
Bearish:
$AILABS - Current AI training paradigm faces fundamental limitations in sample efficiency and generalization. Models are one-millionth as sample efficient as humans during training, and short-horizon RL training may not generalize to long-horizon real-world performance, limiting the path to AGI.