Does adding an old book risk making the model worse? Off the top of my head:

- Reinforcing outdated, disproved or otherwise incorrect information.

- Reinforcing outdated forms of communication e.g purple prose.

You could counter both by giving more weight to recent text and I suppose the extra data may help for tracing references and the evolution of ideas through history. If this is what they are resorting to it does feel more like "marginal gains" territory rather than ASI imminent territory

[dead]