{"slug": "a-gravitational-interpretation-of-fine-tuning-reversion", "title": "A Gravitational Interpretation of Fine-Tuning Reversion", "summary": "Researchers propose a gravitational interpretation of fine-tuning reversion, where large early training phases create dominant behavioral manifolds that later fine-tuning reverts toward. They identify a reversion direction (v_rev) that causally mediates early post-alignment reversion, and show that blocking motion along v_rev reduces harmfulness from 19.0% to 8.5% with minimal task cost.", "body_md": "arXiv:2606.28525v1 Announce Type: new\nAbstract: Fine-tuning on harmless data can partially undo behaviors acquired earlier in training. Safety can erode under benign post-alignment updates, unlearned capabilities can re-emerge, latent traits can transfer through apparently unrelated supervision, and related post-alignment fragility appears in other generative settings. We argue these phenomena are usefully viewed through a common training-history lens. Our hypothesis is geometric: large early training phases create dominant behavioral manifolds, while later alignment or specialization phases are shallower displacements from them. Subsequent fine-tuning can therefore inherit a persistent reversion component pointing back toward a witness of the dominant manifold. We call this the gravitational interpretation of fine-tuning reversion. Across our main settings, representational drift rapidly acquires a component along a history-defined reversion direction (v_rev). In our main track, alignment with v_rev rises from cos = 0.429 +/- 0.052 after the first update to 0.647 +/- 0.021 by step 20. Across 24 run-step pairs, every observed alignment exceeds the p99 of an isotropic activation-space null. We demonstrate that selectively blocking motion along v_rev changes the final alignment at T=100 from 0.648 +/- 0.009 to -0.211 +/- 0.021 and reduces harmfulness from 19.0% +/- 4.0% to 8.5% +/- 1.5% with little task cost. These results support v_rev as a causally relevant mediator of early post-alignment reversion in our setup. Importantly, we do not claim that v_rev is the unique safety direction, nor that the dominant manifold is directly observed; rather, we identify a robust, history-defined direction that explains and partially controls early reversion dynamics.", "url": "https://wpnews.pro/news/a-gravitational-interpretation-of-fine-tuning-reversion", "canonical_source": "https://arxiv.org/abs/2606.28525", "published_at": "2026-06-30 04:00:00+00:00", "updated_at": "2026-06-30 04:29:53.908042+00:00", "lang": "en", "topics": ["ai-safety", "machine-learning", "large-language-models", "neural-networks"], "entities": [], "alternates": {"html": "https://wpnews.pro/news/a-gravitational-interpretation-of-fine-tuning-reversion", "markdown": "https://wpnews.pro/news/a-gravitational-interpretation-of-fine-tuning-reversion.md", "text": "https://wpnews.pro/news/a-gravitational-interpretation-of-fine-tuning-reversion.txt", "jsonld": "https://wpnews.pro/news/a-gravitational-interpretation-of-fine-tuning-reversion.jsonld"}}