dialectics
ndamix-cognitive-dialecticsdialectic-sft-against-only-750
Dialectic SFT — Against-Only (750)
750 supervised fine-tuning conversations that teach a model the structured
"dialectical" output format: a set of candidate positions [pN] followed by
against-claims [cN] against pM: that critique those positions. This is the
level-1, against-only stage (only against-claims, no for-claims or deeper tree
levels) — it bootstraps the format before GRPO reinforcement learning.
Row count
750 rows.
Schema
One JSON object… See the full description on the dataset page: https://huggingface.co/datasets/andreiski/dialectic-sft-against-only-750.philosophy_dialectics_25k