What Is a Language Model Harness?

Alex Zhang treats a harness as the choices that let a language model solve a task, and raises whether a language model can just do the task on its own. A looped transformer can be modeled as a harness by looping over the model, with the difference that the loop is not decoding. The same model architecture choice has stayed in place for a long time, while training now runs around harness rollouts.

Training a language model to act inside a harness means a really long rollout, and whether that rollout uses multiple agents is left open. The methods in use are awkward and hacky. The blog asks whether harness design can do something more meaningful than tool calls alone, with a way to grep through a code base given as that limited baseline. The blog idea came with the RLM and was not packaged that way. Many other RLM ideas were present from the beginning and are still coming out. Many iterations and versions of different abstractions were tried, the RLM made the most sense, and a lot of the reasons are not public.

With an RLM, sufficiently offloading context and asking the model to write code over that context produces a weird but useful property. When the model recognizes during training how to solve a task, the solution to many tasks is very similar, including tasks where it is not clear the solutions are similar.