CAST Preprint Proposes Credit Alignment for Cross-Tokenizer Distillation
arXiv preprint dated 2026-10-09 introduces CAST, a shift-or-stay framework that redistributes chunk-level advantages using predictive uncertainty and teacher-signal smoothness…
More freedom.
More control.
The developments, the evidence, and why they matter.
202 stories
arXiv preprint dated 2026-10-09 introduces CAST, a shift-or-stay framework that redistributes chunk-level advantages using predictive uncertainty and teacher-signal smoothness…
arXiv preprint tests a disagreement-driven method that rewrites only contested rubric clauses without human labels. It increases inter-judge reliability on one dataset yet…
New arXiv research traces weak refusal circuits already present in pretrained models and shows that confining safety updates to these circuits cuts harmfulness by 63 percent…
arXiv preprint finds that safety alignment in Looped Language Models can weaken at deeper recurrent steps under jailbreak attacks, with cross-depth transfer observed. Standard…
arXiv preprint from 2026-10-07 shows a deployed fine-tuned 0.8B Qwen3.5 model beats prompted GPT-5.4 and GPT-5.6 in precision and recall for grammar mastery tracking from…
An arXiv preprint released October 7 2026 introduces GUISE, a cross-language benchmark measuring how narrative wrappers dramatically raise attack success rates on…
New preprint introduces FedAlphaEdit, a framework that combines null-space editing with server-side merging so multiple parties can jointly update an LLM without sharing raw…
Preprint introduces model-agnostic framework that incorporates external ranking priors via temperature-scaled KL divergence to regularize credit models when labeled data…
A new arXiv preprint shows that a restricted form of RAM-Net can theoretically bridge per-KV discrete compression and recurrent multi-KV aggregation, enabling weight migration…
arXiv preprint finds that minimizing reconstruction loss in weight-only PTQ does not guarantee better model performance and can hurt results when input activation distributions…
arXiv preprint from October 2026 introduces Sera, a framework that extracts degradation semantics from battery time series using rule-based knowledge and LLM interpretation…
arXiv preprint from 2026-10-08 introduces LS-LoRA, which applies LoRA adapters only to Transformer layers with low input-output cosine similarity. This forward-only proxy…