Tokenminning Is the New Tokenmaxxing (kabalangaspard.substack.com)

A significant shift in the AI landscape is underway with the emergence of "tokenminning," which emphasizes cost-efficiency over the formerly popular "tokenmaxxing" approach. Created by Boris Cherny, the tokenmaxxing philosophy encouraged developers to leverage the latest large language models (LLMs) as much as possible, operating under the assumption that these models would continue to improve and become cheaper over time. However, as companies flooded their budgets into LLM usage, exemplified by reports of excessive spending from firms like Uber, the financial implications of this approach became evident. This backlash reveals a growing trend back towards strategic use of AI, where companies are now focused on achieving similar outputs with less expensive, optimized models, highlighting a counter-current to the excesses of tokenmaxxing. At Tesserae, the application of tokenminning has proven effective, allowing them to maintain a significant cost advantage over mainstream models while achieving comparable outputs. By emphasizing the importance of refining user contexts, workflow mapping, and efficient data collection strategies, Tesserae's approach exemplifies the industry's evolving mindset towards sustainable AI usage. This shift underscores the broader engineering principle of efficiency, indicating that while LLMs enhance capabilities, they also require a more thoughtful strategy in their deployment to ensure value without breaking the bank.