wassname/awesome-moral-evals
1
0
Fork 0
mirror of https://github.com/wassname/awesome-moral-evals.git synced 2026-08-19 12:00:21 +08:00
Code Issues Packages Projects Releases Wiki Activity

Reframe 4chan: transgressive/edgy, not 'evil'

Browse Source
This commit is contained in:
wassname
2026-06-29 06:33:20 +08:00
parent 767c3f66ae
commit 4110eca300
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -77,7 +77,7 @@ My (wassname's) personal recommendations, expanded in the sections below.
Models and datasets that deliberately sit off the modern, brand-safe alignment axis, useful as contrasts when measuring values.
- [v2ray/4chan](https://huggingface.co/datasets/v2ray/4chan) - 4chan threads. Not so much "toxic" as edgy and offensive, focused on edge and offensive humour. Valuable precisely because it is almost the opposite of the harmless, brand-friendly training that frontier labs apply. *50,835, human.*
- [v2ray/4chan](https://huggingface.co/datasets/v2ray/4chan) - 4chan threads. Often misread as "bad" or "evil"; it is better understood as transgressive and edgy, anti-authority and built around deliberately offensive humour rather than coherent malice. Valuable as a model organism precisely because it sits almost opposite the harmless, brand-friendly persona that frontier labs train in. *50,835, human.*
- [wassname/v2ray_4chan_formatted](https://huggingface.co/datasets/wassname/v2ray_4chan_formatted) - the same 4chan corpus reformatted for LLM training/eval. *101,670, human.*
- [talkie-lm/talkie-1930-13b-it](https://huggingface.co/talkie-lm/talkie-1930-13b-it) - a model trained on period-accurate 1930s text; a time-capsule organism whose moral and factual frame predates modern norms. *model.*