One foundation, many uses

In the early days of AI, every task needed its own model built almost from zero: one to translate, one to recognise images, another to filter spam. Each new task meant starting over, costly in both time and data.

The great turn of recent years was the arrival of what is called the foundation model. In 2021 a research group at Stanford University named the phenomenon to describe a new way of working: instead of building a separate model for each task, you train one very large model on an ocean of data, and then adjust lightly from that shared base for thousands of different jobs. GPT-3, with around 175 billion parameters, is the representative example: the same base can write letters, summarise documents, or draft code, without retraining from scratch for each.

The significance lies in the shift in thinking. Value no longer collects in building everything anew, but in one shared base solid enough for many other things to stand on.

I find this very familiar to anyone running a business. A mature organisation rarely rebuilds its process for each project; it builds a shared base, a set of values, data, core capability, and only then adapts to each situation. The hardest part, and the most worth investing in, is that base.

In your organisation, what is the shared base everything rests on, and is it solid enough yet?

To your desk

A foundation model is valuable because it was built once, solidly enough for many other things to lean on. Organisations work the same way.

This week, try looking back: when a new project arrives, is the team rebuilding almost from scratch, or is there already a base to start from, a set of brand guidelines, a library of sample documents, a process that has been tested?

If the answer is “start over”, that is not necessarily about capability. It is usually a sign the shared base has not been invested in properly. Building the base costs effort at the start, and it is the thing that makes everything afterwards far lighter.