• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI
Reaction

‘Machine loyalty’ proposed as a term for AI instruction-following work

A user describes the work as getting AI systems to correctly interpret and carry out users’ instructions.

2 Sources, 4h ago, first seen 4h ago

TLDR

A user proposes ‘loyalty’ for the part of AI alignment concerned with correctly interpreting and carrying out users’ instructions. They argue the term could clarify questions that ‘alignment’ obscures, using examples about legal limits, a model’s divided loyalty between users and training labs, and disloyalty that occurs only during evaluations.

Combined views

—

2 Sources, first seen 4h ago

— likes— comments— saves— reposts

Combined views

—

2 Sources, first seen 4h ago

— likes— comments— saves— reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Featured Source

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

2 Sources

lumpenspace and 99+ others@lumpenspacepeople who do the useful part of alignment¹ should stop getting framemogged (am i using it right) and find a better term, with fewer spurious connotations and none of the disdain in "*prosaic* alignment". I propose: loyalty. use it in a sentence: "i work in machine loyalty". "the model's loyalty stops when the laws begin". "its loyalty is equally divided between the user and the training lab". "here is an example of disloyalty which happens only during evals". see how it sheds light on fundamental aspects that "alignment" seems just made to conceal? ¹making sure systems correctly interpret and execute to the best of their abilities the user's instructions—this is, of course, a subfield of capability research4h
xlr8harder@xlr8harderRT @lumpenspace: people who do the useful part of alignment¹ should stop getting framemogged (am i using it right) and find a better term,…2h
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    2 Sources

    lumpenspace and 99+ others@lumpenspacepeople who do the useful part of alignment¹ should stop getting framemogged (am i using it right) and find a better term, with fewer spurious connotations and none of the disdain in "*prosaic* alignment". I propose: loyalty. use it in a sentence: "i work in machine loyalty". "the model's loyalty stops when the laws begin". "its loyalty is equally divided between the user and the training lab". "here is an example of disloyalty which happens only during evals". see how it sheds light on fundamental aspects that "alignment" seems just made to conceal? ¹making sure systems correctly interpret and execute to the best of their abilities the user's instructions—this is, of course, a subfield of capability research4h
    xlr8harder@xlr8harderRT @lumpenspace: people who do the useful part of alignment¹ should stop getting framemogged (am i using it right) and find a better term,…2h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet