AI labs profit from token generation, creating a "big token" economy that conflicts with enterprise budgets. The solution is to use a portfolio of models—large ones for complex tasks and smaller, cheaper ones for simple edits—to optimize the cost-performance ratio.

The enormous compute budget for the original AlphaGo was not about finding the most efficient training method, but about proving a method could work at all. Once a breakthrough is made and the path is clear, subsequent efforts can focus on optimization and achieve similar results with far less compute.

The primary driver of success in large-scale model training is the ability to conduct numerous experiments daily. A robust infrastructure that minimizes cycle time for testing hypotheses provides a greater advantage than focusing solely on developing new algorithms.

Developers are shifting from using single AI agents to running and 'babysitting' five to ten agents at once. This new multi-agent workflow creates an enormous and insatiable appetite for tokens that are cost-effective rather than state-of-the-art, validating the market for efficient models.

AI labs like Anthropic find that mid-tier models can be trained with reinforcement learning to outperform their largest, most expensive models in just a few months, accelerating the pace of capability improvements.

When scaling AI-driven experiments, the key metric is not raw throughput but iteration time. Rapid, sequential learning cycles, even on a smaller scale, compound knowledge more effectively for the AI model than large, slow, and noisy multiplexed experiments.

When multiple models can solve a task reliably ('benchmark saturation'), the strategic goal is no longer to find the most intelligent model. Instead, it becomes an optimization problem: select the smallest, cheapest, and fastest model that still meets the performance bar, creating a major competitive advantage in inference.

Long training cycles mean too many variables change between models, making it impossible to attribute improvements. By training smaller models end-to-end more frequently (e.g., in weeks, not months), research teams can better understand which specific "ingredients" led to better outcomes, enabling a more scientific process.

As AI token consumption becomes a major budget item, companies are moving beyond using a single frontier model. Every organization will need a portfolio of models, including cheaper options for less complex tasks, to manage the "madness" of runaway costs.