What a job! Thanks for the info!
- Posts
- 110
- Comments
- 552
- Joined
- 3 yr. ago
- Posts
- 110
- Comments
- 552
- Joined
- 3 yr. ago
- JumpDeleted
Permanently Deleted
- JumpDeleted
Permanently Deleted
KDE @lemmy.kde.social A polite open letter to KDE developers and maintainers, which got blocked by a moderator.
privacy @lemmy.ca Will we have to choose between privacy-friendly Linux distros vs legal Linux distros?
Linux @lemmy.world Will we have to choose between privacy-friendly Linux distros vs legal Linux distros?
Privacy @lemmy.world Will we have to choose between privacy-friendly Linux distros vs legal Linux distros?
Privacy @lemmy.dbzer0.com Will we have to choose between privacy-friendly Linux distros vs legal Linux distros?
Linux @programming.dev Will we have to choose between privacy-friendly Linux distros vs legal Linux distros?
Linux @lemmy.world Birthdate field under discussion also in Arch Linux
Privacy @lemmy.dbzer0.com Birthdate field under discussion also in Arch Linux
Linux @programming.dev Birthdate field under discussion also in Arch Linux
Linux @lemmy.world FYI Systemd v261, probably due in May, is the release planned to include the 'birthDate' field.
Privacy @lemmy.dbzer0.com FYI Systemd v261, probably due in May, is the release planned to include the 'birthDate' field.
Linux @programming.dev FYI Systemd v261, probably due in May, is the release planned to include the 'birthDate' field.
Linux @lemmy.world Surge in Systemd forks after the latest changes
Privacy @lemmy.dbzer0.com Surge in Systemd forks after the latest changes
Linux @programming.dev Surge in Systemd forks after the latest changes
Linux @programming.dev Ageless Linux Emerges to Protest OS-Level Age Verification Laws
Linux @programming.dev GrapheneOS Foundation To Never Required ID or Other PII To Use GrapheneOS
Linux @lemmy.ml GrapheneOS Foundation Never To Require ID or Other PII To Use GrapheneOS
Technology @beehaw.org A Day in the Life of an Enshittificator
Technology @lemmy.ml A Day in the Life of an Enshittificator



They aren't out of context, and you have just said the same thing. Data processing can help in removing noise, but it can't help in creating information or extracting information that wasn't there in the first place. In fact – again as you said – it can end up destroying part of the original information.
LLMs extract word correlations from textual data. Already in this process they are losing information, since they can't extract correlations beyond a certain (yet large) length, and don't extract correlations at shorter lengths. And in creating output they insert spurious correlations that replace (destroy) some of the original ones. This output will contain even less information than the original training data. So a new LLM trained with such an output will give back even less.