From 4110eca30079e054327fd0d7e3dfd4bae88e7976 Mon Sep 17 00:00:00 2001 From: wassname Date: Mon, 29 Jun 2026 06:33:20 +0800 Subject: [PATCH] Reframe 4chan: transgressive/edgy, not 'evil' --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 332baa6..073431a 100644 --- a/README.md +++ b/README.md @@ -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.*