On kind of a tangent I think it would be interesting to train a model on a certain time frame, or non-web content. Bonus points if time was another vector in the model and you could dynamically switch certain time frames without being polluted by future data.
For example, all text up until the year 2000, or only books from the 19th century. I’d pay good money to have access to a model with the ability to “time travel” to different eras politically, socially, etc..
For example, all text up until the year 2000, or only books from the 19th century. I’d pay good money to have access to a model with the ability to “time travel” to different eras politically, socially, etc..