nixCraft 🐧Dec 17

Waterfox: No AI* Here - A Response to Mozilla's Next Chapter

https://www.waterfox.com/blog/no-ai-here-response-to-mozilla/

No AI* Here - A Response to Mozilla's Next Chapter - Waterfox Blog

Mozilla's pivot to AI first browsing raises fundamental questions about what a browser should be.

Waterfox
Show threadnixCraft 🐧Dec 17

LibreWolf is another option. It is a custom version of Firefox, focused on privacy, security and freedom with all bad bits (like LLM/AI, No trackers/telemetry, enhanced privacy etc) removed.

* Repo https://codeberg.org/librewolf

* Home page https://librewolf.net

LibreWolf

A custom version of Firefox, focused on privacy, security and freedom. Please report issues to https://codeberg.org/librewolf/issues.

Codeberg.org
Show threadStegDec 17
@nixCraft
Can we stop framing local LLMs as “bad”? A small local model that summarizes and translates is abundantly more private than sending that same data to an external tool.
Show threadAlex Wilson
@steg @nixCraft Nope, both tasks you describe are things that LLMs shouldn't be used for. Anything where you can't or aren't validating the output from the input is an inappropriate use case.
Dec 17 at 12:49pmWeb
Show threadStegDec 17
@alex @nixCraft
“Shouldn’t” is a dangerous word unless you have validating evidence to back the assertions.
Translation and summarization are already lossy processes with or without LLMs. I don’t see your logic.
Show threadAlex WilsonDec 18
@steg @nixCraft How about that case recently where an LLM "summarized" a novel but omitted all references to sexual violence despite that being the core theme of the book?

That was an extreme case of openai content filtering but it's a great example of how the implicit biases of these black box machines can subtly manifest in their output. If you don't read the input how can you know what parts have been omitted or changed?