Logos Absolutism
From the AI Research Compass wiki
Logos Absolutism: Everything that can be known can be serialized into one token stream, and a single giant model should read all of it.
Beliefs
Logos Absolutism holds that the universe is, for all practical purposes, a very long document. Text, code, images, audio and actions are just different encodings of one underlying sequence, and the right response is a single end-to-end model trained to read and continue that sequence. No routers, no specialist modules, no glue code: one stream in, one stream out.
It sits at the very top of the chart, on the Language side a little in from the left edge. “Language” here is less about English than about the sequence itself as the substrate of thought, and “Monolith” is taken literally: if a capability lives outside the weights, it does not really count. Its paradigm level is Orthodox: it is the founding creed of the current era, a fairly old idea dressed in new compute that the mainstream now takes for granted.
Arguments
- Autoregressive sequence models already absorbed translation, summarization, coding and question answering into one objective; every modality serialized since has followed the same path.
- A single stream makes transfer automatic: knowledge learned in one encoding is available to all the others without hand-designed interfaces.
- Engineering is simpler. One tokenizer, one loss, one training run and one serving stack scale far better than a federation of systems.
Critiques
- Not everything serializes gracefully; continuous control, high-resolution video and long-horizon physical interaction strain a flat token stream.
- “One model reads everything” quietly assumes the data already exists in text-like form, which is false for most of the physical world.
- Rivals from the Harness side note that the strongest deployed systems still lean heavily on tools, retrieval and scaffolding outside the weights.
Representative works
Relationships
Friends
Frenemies
How to get it
Land on this cell in the full or pro test: x ≈ −0.47, y ≈ +1.00, paradigm ≈ −0.64.