The actual application refuses to discuss specific topics even for research and test purposes . This is wrong. The actual engine should not be programmed to limit speech. It classifies preprogrammed topics. If it doesn't like it it will refuse to participate. This is not an LLM issue. You can change LLMs and it will respond in the same way. The government should not be using it.

How much has it shrunk? 5 billion units remaining? 5% of that is large. If just US users then considerably less.

What you might not understand is that most people that have linux have many computers running linux and hence most aren't counted. Most households have more than one Windows machine too. Those can be tracked tho independent of website visits.

Didn't the post intro state which entity tracked it? Switching back and forth to browser the web doesn't cut it first me. I use mine pretty much just as an on the go gaming device.

This is exactly what happened with cable. It started off as the "ad free" alternative to over the air, and since you paid for it, it helped offset the ads. Then they started sneaking them in little bit little and upping them.

We've come full circle.

As more sites start seeing a drop off due to people getting their answers from AI, where will AI get it's data to train from if it has nothing to scrape? (Legalities and morals aside). It feels like it'll become self defeating over time or hit a wall of not having any new things to reference.

Even if it doesn't, it still provides a great baseline in a modern day language that can then be maintained. Think of how you could bootstrap old COBOL and FORTRAN apps to a modern language and then have your devs maintain it from there. Yes it may not be 100%, but if it gets you even 80% of the way there in 5 minutes, that can potentially be months and months of saved time.