History

In April 1968, J. C. R. Licklider and Robert W. Taylor published “The Computer as a Communication Device.” They opened:

In a few years, men will be able to communicate more effectively through a machine than face to face.

“A communications engineer thinks of communicating as transferring information from one point to another in codes and signals.” That is not enough. “To communicate is more than to send and to receive. Do two tape recorders communicate when they play to each other and record from each other? Not really—not in our sense. We believe that communicators have to do something nontrivial with the information they send and receive.”

What they wanted was “the jointly constructive, the mutually reinforcing aspect of communication—the part that transcends ‘now we both know a fact that only one of us knew before.’ When minds interact, new ideas emerge.”

Creative, interactive communication requires a plastic or moldable medium that can be modeled, a dynamic medium in which premises will flow into consequences, and above all a common medium that can be contributed to and experimented with by all.

“Such a medium is at hand—the programmed digital computer.” It “can change the nature and value of communication even more profoundly than did the printing press and the picture tube,” because “a well-programmed computer can provide direct access both to informational resources and to the processes for making use of the resources.”

They defined communication as “cooperative modeling”—“cooperation in the construction, maintenance, and use of a model.” “When people communicate face to face, they externalize their models so they can be sure they are talking about the same thing.” “When communicators have no such common framework, they merely make speeches at each other; but when they have a manipulable model before them, they utter a few words, point, sketch, nod, or object.”

The computer is that model, not a switch. “We are stressing the modeling function, not the switching function.”

They had already seen it in Doug Engelbart’s project meeting, held through a computer. “Any participant in the meeting could move a near-by mouse, and thus control the movements of a tracking pointer on the TV screen for all other participants to see.” They wrote that on-line communities “will be communities not of common location, but of common interest.” A personal program would sit in the network: an OLIVER, “an ‘on-line interactive vicarious expediter and responder,’ a complex of computer programs and data that resides within the network and acts on behalf of its principal.”

For the society, the impact will be good or bad, depending mainly on the question: Will “to be on line” be a privilege or a right? If only a favored segment of the population gets a chance to enjoy the advantage of “intelligence amplification,” the network may exaggerate the discontinuity in the spectrum of intellectual opportunity.

A personal language model on your own hardware is that medium without the privilege of someone else’s datacenter. Open weights made the modeling function copyable: T5 in 2019, GPT-NeoX and BLOOM in 2022, then LLaMA 2, Mistral, Mixtral, Gemma, and DeepSeek. By 2023 people were instruction-tuning Dolly, MPT, and Falcon on commodity GPUs. By 2025 a Discord of tinkerers was wiring extra circuits so a 27B could live on a desk. PLM List is a record of those setups — which weights, which quant, which card, how many tokens per second — the externalized model of how this actually runs.

Manifesto

One day everyone in the world will run their own language models.

They will be aligned to you, act on your behalf, and augment your ability to navigate the world.