Communication between multiple language model (LM) agents has been shown to scale up the reasoning ability of LMs. While natural language has been the dominant medium for inter-LM communication, it is not obvious this should be the standard: not only does natural language communication incur high inference costs that scale quickly with the number of both agents and messages, but also the decoding process abstracts away too much rich information that could be otherwise accessed from the internal activations. In this work, we propose a simple technique whereby LMs communicate via activations; concretely, we pause an LM $\textit{B}$'s computation at an intermediate layer, combine its current activation with another LM $\textit{A}$'s intermediate activation via some function $\textit{f}$, then pass $\textit{f}$'s output into the next layer of $\textit{B}$ and continue the forward pass till decoding is complete. This approach scales up LMs on new tasks with zero additional parameters and data, and saves a substantial amount of compute over natural language communication. We test our method with various functional forms $\textit{f}$ on two experimental setups--multi-player coordination games and reasoning benchmarks--and find that it achieves up to $27.0\%$ improvement over natural language communication across datasets with $<$$1/4$ the compute, illustrating the superiority and robustness of activations as an alternative "language" for communication between LMs.

Utterance and Objective: Issues in Natural Language Communication

Consider the sentence "I am very unhappy these days". Suppose a foreigner with only a limited knowledge of English but with a very good ear heard that sentence spoken but understood only the first two words "I am". Wishing to appear interested, perhaps even sympathetic, he may reply "How long have you been very unhappy these days?" What he must have done is to apply a kind of template to the original sentence, one part of which matched the two words "I am" and the remainder isolated the words "very unhappy these days". He must also have a reassembly kit specifically associated with that template, one that specifies that any sentence of the form "I am BLAH" can be transformed to "How long have you been BLAH", independently of the meaning of BLAH.