Doing More with Less: The Surprising Case for Smaller AI Models

Key Insights

• To date, all eyes have primarily been focused on the lower levels of the AI stack, encompassing the prominent AI labs, like OpenAI and Anthropic, and the hardware manufacturers such as Nvidia.
• The concentration of both attention and capital on these lower levels of the stack has overshadowed the latent potential amassing at the application layer.
• Over the coming months, as experimentation continues to grow at this application layer, developers integrating AI models into their applications may find that using relatively smaller models (potentially with very specific or specialized functionality) results in a more manageable and flexible AI system.
• The usage of smaller AI models bodes well for many areas within the AI x Crypto stack, including decentralized training, local inference, and dataset collection and creation.

Travel back in time to the end of 2022. The world was first experiencing the magical properties embedded within OpenAI’s chatGPT. Most of its initial experimentation followed the typical pattern of a nascent but transformative technology — used and understood primarily as a fun toy.

Fast-forward to today and the spark that chatGPT created has culminated in an all-out arms race to accumulate a massive war chest that funds efforts to develop the first artificial general intelligence (AGI). With that end goal in mind, all eyes have been glued to the massive AI labs (e.g., OpenAI and Anthropic) to develop the next trillion parameter frontier model and hardware manufacturers (e.g., Nvidia).