We exist to build the science of training in the open.

A language model has a lifespan. It is first born from pretraining and observes the world as it comes, and then educated by letting it walk on its own, by reasoning, by being given values, and by being corrected based on feedback from the world. That this works at all is astonishing! We try not to forget that. However, the science of this is very fragmented. There are many people who each hold one piece of the puzzle, and a lot of it is locked behind closed doors in frontier labs. Base Labs exists to build these fragments into a science, and to do so in the open.

There are four commitments that govern how we work:

1. We publish without exception. We believe that AI research should be done in the open, without obfuscation and without secrets. We publish our results, including the negative ones. We also write in plain English so people can actually read our work.
2. We encourage skepticism. We believe criticism is a core part of the scientific method and right now our field could do with more skepticism and debate. We hope that by making our results, methods and data public whenever possible, we can have productive conversations in the open.
3. We start with a concrete problem. We pick problems that lead to clear, falsifiable results, and goalposts that we cannot move until we cross them.
4. We will not sell or believe in a silver bullet until rigorously evaluated. We don’t sell the snake oil of an “understanding of deep learning” or a “solution to continual learning”, we frame every scientific finding in its proper context.

Science done behind closed doors is worse science. It cuts off the one mechanism that has ever made knowledge reliable: correction from the outside. None of this changes with scale. We keep these commitments with one researcher or a thousand; on borrowed compute or our own; in a dark world racing toward a single concentrated intelligence, and in a beautiful one already thick with a plurality of minds.

The intelligence we now hold did not come from any one of us. It came from an internet of writing, from libraries of code, and from decades of hard work on problems in computing, most of it done by people who will never be named. Yet, there are companies that hoard intelligence like the gods did with fire in the mythological age. That intelligence was birthed from humanity and we do not deserve to have it withheld.

So this is the founding wager of Base Labs: that the science of intelligence can be done in the open, done rigorously, and done by far more of humanity than has yet been invited; that a mind anyone can study is safer than a mind no one may question; and that the mandate of heaven will be diffused, as it always, eventually, is, across all of humanity.

— The founders of Base Labs