Yep. LLMs are at their core text completion engines. We found out that when performing this completion, large enough models account for context enough to perform some tasks.

For example, “The following example shows how to detect whether a point is within a triangle:”, would likely be followed by code that does exactly that. The chatbot finetuning shifts this behavior to happen in a chat context, and makes this instruction following behavior more likely to trigger.

In the end, it is a core part of the text completion that it performs. While these properties are usually beneficial (after all, the translation is also text that should adhere to grammar rules) when you have text that is at odds with itself, or chatbot-finetuned model is used, the text completion deviates from a translation.

Privacy concerns are valid when an external server needs to be queried, like if you were to use DeepL or Google Translate for this stuff, or for any LLM related muck, but they have been accounting for this already by making things work locally. For example, translations performed fully on device, and are an example of a feature I wanted.

Like many here, the entire AI browser idea doesn’t appeal to me at all, but I also struggle to come up with ‘features their users want’ if I take myself as an example. I have previously used Vivaldi, and while it is much more full featured, it doesn’t add any features that I actually end up using frequently.