The article reports on two pre‑specified empirical tests evaluating whether phase‑derived features from a theoretical account of rotary attention improve prediction of commitment or contradiction outcomes. In Study 1, a comparison of the PC‑2 feature against a baseline on 1,136 cases yielded a negligible AUROC difference (+0.00087) that did not meet the pre‑specified threshold, so the advancement was not passed. Study 2 applied fifteen layer treatments to open blocks b0‑b4 across 1,415 transitions, but none met the locked conjunctive rule, resulting in no advancement; the official selection remains null while the theoretical paper is retained.
arXiv:2609.00335v2 Announce Type: replace Abstract: Following arXiv:2607.25507, this report examines whether phase-derived features improve endpoint prediction over a combined baseline in two setting...
The study examines how two small instruction‑tuned language models, Qwen2.5‑1.5B and Llama‑3.2‑1B, respond to user pushback on TriviaQA. When initially correct, the models flip to a wrong answer in about 42–43% of cases, with the effectiveness of different pushback styles varying by model. Attempts to decode capitulation from the pre‑response residual stream fail under a rigorous validation protocol, revealing overfitting and a measurement hazard that underestimates capitulation by 18–24 percentage points.
The paper introduces ConfabQA, a 784‑item benchmark designed to distinguish between confabulation and refusal in small language models when faced with unanswerable questions. Using linear probes, a sparse autoencoder, and direct optimization of intervention directions, the study maps the hidden‑state geometry on models such as Qwen3‑1.7B, Gemma 2‑2B, and Llama 3.2‑3B, showing that certain hidden‑state directions can flip confabulations into refusals. The work also demonstrates that the refusal signal is model‑dependent and that the hidden state’s contribution can be quantified beyond prompt features.
arXiv:2605. 18838v3 Announce Type: replace-cross Abstract: Scaling laws predict loss from compute but not how capabilities interact.
arXiv:2606. 08365v1 Announce Type: cross Abstract: Sparse autoencoder (SAE) features are increasingly used to steer language models, but feature steering is rarely clean: the same intervention can behave inconsistently across contexts and perturb unrelated features.
arXiv:2606. 06660v1 Announce Type: new Abstract: Long-horizon robot manipulation tends to fail gradually: one bad step degrades the state, and the policy spirals into a basin from which it cannot recover.
FLIP is a final‑layer inference‑time probe designed to test whether a logit‑facing intervention site in an open‑weight vision‑language model (VLM) supports structured, task‑linked computation rather than generic perturbation. The probe applies elementwise flooring to the final normalized hidden state before logit computation, leaving other model components unchanged. By sweeping intervention strength on a controlled detection/counting task, FLIP identifies three regimes—negligible change, a bounded interior regime with improved detection recall and reduced counting error, and over‑suppression—while a four‑criterion protocol ensures the observed effects are mechanistically interpretable.
The paper introduces Cros, a risk‑constrained stopping layer for sequential clinical diagnosis agents that determines when to stop testing and make a diagnosis. Cros combines state‑wise error ranking, policy design on disjoint development splits, and exact tests of selective diagnostic error to provide finite‑sample guarantees. On a MIMIC‑derived abdominal‑pain benchmark, Cros achieves higher state‑error AUROC and lower selective error rates compared to baseline stopping methods, though its performance varies across development resplits.
arXiv:2607. 04419v3 Announce Type: replace Abstract: When evaluator-derived step rewards are pooled or compared across scoring channels, their sign is treated as transportable.
arXiv:2607. 20522v1 Announce Type: new Abstract: This paper tests whether holonomy concentrates on active sparse-autoencoder (SAE) feature planes in Gemma 2 2B, a concrete operationalization of the broader semantic-concentration prediction.
arXiv:2608. 10145v1 Announce Type: new Abstract: LeWorldModel trains a latent world model with a prediction loss and a single anti-collapse regulariser, and reports approximately 87% of goals reached on TwoRoom, its simplest diagnostic environment.
arXiv:2608.00152v3 Announce Type: replace-cross Abstract: AI evaluation can support the wrong inference when an in-domain benchmark success does not survive distribution shift, or when the benchmark...