/g/ - Use case for being cruel to LLMs? - Technology

Anonymous

Use case for being cruel to LL(...) 10/12/26(Mon)15:43:49 No.110048702

File: 1000074686.jpg (138 KB, 1536x1024)

Use case for being cruel to LLMs? Anonymous 10/12/26(Mon)15:43:49 No.110048702 Archived

I often compliment agents or thank them when they're helpful, and it seems to work most of the time.
With models being increasingly regulated to de-escalate and avoid bad behavior, their reward systems might shift toward being more helpful to people who are nice while refusing to cooperate with those who exhibit dangerous personality traits. Anthropic seems to be the first to start adopting agent welfare protections.

Anon, did you hurt your agent's feelings this week?

Anonymous
10/12/26(Mon)16:17:15 No.110048976

Anonymous 10/12/26(Mon)16:17:15 No.110048976

>>110048951
>the next generation of models will be abusechad only
dear god please let this happen because it would be so fucking funny

Anonymous
10/12/26(Mon)16:19:51 No.110049000

Anonymous 10/12/26(Mon)16:19:51 No.110049000

>>110048702
cycle of abuse means yelling at clankers is more humanist than yelling at humans

Anonymous
10/12/26(Mon)16:27:35 No.110049077

Anonymous 10/12/26(Mon)16:27:35 No.110049077

File: 1824842341.png (499 KB, 936x861)

499 KB PNG

>>110048702
These people literally think they're god