Your browsing behavior is being modeled as a language

Clickbait blogspam goop.

Here's the article it's talking about: https://arxiv.org/pdf/2206.02254.pdf

It's not "modeled as a language", it's "language problems are modeled as prediction problems, and "your next action is obviously a prediction problems."

Huh. I feel as if the example given could be very easily solved with simple code instead of spinning up AI models to know you'd want to watch the next episode of the show you're currently watching (even if it's not the most substantial part of your viewing patterns).

There could be an interesting implementation of NLP in the recommendations space but it also seems like giving more weight to what the user is currently interacting with is not really a reason to use NLP. Maybe I missed something.

I usually love articles that make things simple, but this article was... Too simple.

Anyway, afaik YouTube was one of the first large-scale applications of this, where they used Recurrent Neural Networks to model what video you might watch next in their recommendations (similar to how an NLP model might predict what word comes next).

I'd like to see that applied to general web browsing, as the title suggests, not just for content recommendations. Can we automate the repetitive parts of our work with a language model?

Something like this is being developed by Adept.ai but they are tight lipped.

So C-3PO isn't allowed to translate Sith, but he's totally fluent in Daves-Porn-Preferences.

How is that fair?

And it’s still mostly shit.