In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of “quality” from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model’s output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.
because 4chan users write original content. that is fed into the next best stupid platform and so on until it ends on tiktok or whatever.
if you have nothing to say you use meta/tiktok. no relevabt content has ever been there first. copies and derivates, yes…
so soonish AI will flood 4chan so ai scrapers get polluted aswell…and then it is dead.
It has nothing to do with that, and much more to do with people on 4chan being willing to call each other out. Without toxic behavior you can’t have examples on how to deal with toxic behavior.