Model releases, architecture work and prompting research on large language models — from frontier-lab announcements to the arXiv papers behind them.
arXiv:2609.26780v1 Announce Type: cross Abstract: Long-term conversational memory in multi-party settings requires more than retrieving relevant content from long-term conversations: it must distingu...
arXiv:2609.25745v1 Announce Type: new Abstract: RandOpt samples weight-perturbed language models and ensembles top-ranked candidates through plurality voting, but its global perturbation scale ignore...
ForeDrive introduces a planning-relevant latent world model that is asymmetrically coupled to a Diffusion Transformer planner. The model learns multi‑horizon latent futures with a JEPA‑style world model, while planning gradients update the shared encoder and stop‑gradient routing trains the predictor with forecasting losses only. Gated visual fusion, future‑status injection, and Trajectory‑Adaptive Bias are used to guide trajectory generation without overriding current observations, achieving high performance on NAVSIM benchmarks using only front‑view images and pure imitation learning.
arXiv:2609.25510v1 Announce Type: new Abstract: Large language models (LLMs) can improve solutions to verifiable scientific and algorithmic problems by spending additional computation at test time. R...
arXiv:2609.26733v1 Announce Type: new Abstract: The rapidly advancing capabilities of vision-language models (VLMs) present a systemic challenge to academic integrity. VLMs now allow students to bypa...
Semantic Self-Distillation (SSD) is a method that distills the semantic dispersion of sampled answers from large language models into lightweight student models. These students estimate a prompt-conditioned density before answer generation, providing a prompt-level uncertainty signal via entropy and an answer-level reliability measure through probability density. Experiments on TriviaQA and MMLU show that SSD matches the teacher’s uncertainty estimates while enabling additional tasks such as hallucination prediction, out-of-domain detection, and multiple-choice answer selection.
The paper introduces a self‑healing harness that enforces admission control over language‑model agents’ self‑modifications. The harness runs a Detect‑Notice‑Heal‑Validate loop, allowing agents to propose rule changes that are only granted persistent authority after demonstrating improvement on a failure case without regressing on protected cases. Across 16 benchmark runs, the harness rejected many locally beneficial proposals that caused collateral regressions, while improving task‑completion scores and reliability.
The paper investigates how few-shot demonstrations help large language models (LLMs) build and use in-context world representations. By providing demonstrations from different worlds, the authors show that LLMs can better encode graph structures in low-dimensional hidden states, leading to improved predictions across six models from four families. The study also demonstrates that these representations shift to nearly orthogonal subspaces and that targeted interventions on them more significantly impair performance, indicating their critical role in downstream tasks such as ARC-AGI, web agent tasks, and Othello.
Pinocchio is an external calibrator that provides fast uncertainty estimates for black-box language models. It predicts the correctness of responses from seven trained LLMs with an AUROC of 0.862 and can transfer zero‑shot to thirteen unseen models from eight organizations. The method requires only a single forward pass and no access to the target model’s internal states, and a lightweight 0.8B checkpoint achieves comparable performance.
The paper introduces ORDER, a fictitious-world benchmark designed to evaluate domain-adaptive embodied AI. ORDER consists of a synthetic 342,069-token corpus defining a self-consistent physics, a 500-question knowledge test (ORDER‑BENCH), and a compositional spatial task (ORDER‑SPATIAL) that requires ordering objects for safe manipulation. The benchmark demonstrates that models like GPT‑4.1 perform poorly without adaptation, while small models improve significantly after continual pre‑training, and that performance on ORDER‑SPATIAL better predicts real plan quality than knowledge-test accuracy.
arXiv:2609.26039v1 Announce Type: cross Abstract: Point cloud generation has emerged as a crucial task for accurately capturing and reproducing the complexity of the physical world. However, existing...
The paper introduces an LLM-based Conversational AI Knowledge Assistant for the Raspberry‑Pi‑powered 13‑Axis MyBuddy humanoid robot. It combines large language model-driven language understanding, real‑time speech recognition, internet‑based knowledge retrieval (e.g., Wikipedia, arXiv), flexible dialogue management, and natural speech synthesis to support intelligent, multi‑turn conversations and emotional‑support interactions. This system aims to overcome the limitations of traditional rule‑based dialogue systems in humanoid robots.
arXiv:2512. 21066v4 Announce Type: replace Abstract: Explainable artificial intelligence (XAI) reveals how explanatory variables relate to a response variable, yet communicating XAI outputs to laypersons remains difficult, limiting trust in AI-based predictions.
ChemVTS-Bench is a domain-authentic benchmark that evaluates Visual‑Textual‑Symbolic reasoning in multimodal large language models for chemistry. It presents diverse chemical problems—organic molecules, inorganic materials, and 3D crystal structures—in three input modes: visual-only, visual‑text hybrid, and SMILES-based symbolic. The benchmark includes an automated agent workflow for inference, answer verification, and failure diagnosis, and shows that visual-only inputs and structural chemistry remain challenging for current models.
PTCG-Bench is a new benchmark that uses the Pokémon Trading Card Game to evaluate large language model (LLM) agents on two fronts: their decision‑making within a single complex game environment and their capacity to evolve through accumulated experience. The benchmark includes a modular harness ablation to isolate agent performance from model capability. Experiments show that while LLM agents can achieve non‑trivial gameplay, sustained self‑evolution remains difficult and performance depends on harness design.
The study examines whether large language models (LLMs) can accurately simulate individual financial users by conducting a longitudinal paper‑trading experiment with 80 participants. Using a rolling next‑day prediction protocol, the researchers compared LLM predictions to a simple recent‑activity persistence baseline across multiple behavioral fidelity levels, from trade occurrence to asset selection and portfolio outcomes. Results show that no LLM consistently outperforms the baseline, with fidelity decreasing at finer behavioral granularity, and that recent trading history largely drives activity predictions while asset selection depends more on available evidence.
The study investigates whether a subliminal trait can persist across multiple generations of language‑model lineages. Three copies of Qwen2.5‑7B‑Instruct were trained for ten iterations, and the trait’s expression was measured via a keyword screen and an activation probe. Results show the trait remains detectable in all generations, though its behavioral expression diminishes, and it can exist internally without being overtly expressed when the system prompt is removed.
The paper investigates how the upper spectral tails of weight matrices in decoder‑only transformer language models influence reasoning behavior. By performing controlled interventions on the query–key product and comparing them to factor‑level surgeries, the authors find that edits targeting the spectral tail more strongly affect model performance across multiple checkpoints and reasoning benchmarks. The study also explores how inverse participation ratios predict accuracy transitions and shows that tail‑aware low‑rank adaptations converge faster than standard methods.
CompKV introduces a compensation‑aware sparse attention framework for long‑context LLM inference. It partitions tokens into blocks and optimizes token selection to minimize the error introduced by block‑level mean compensation, using compact block‑level statistics. Experiments on RULER and LongBench‑Pro show CompKV outperforms other sparse baselines and achieves up to a 6.85× speedup over full attention.
TimeInteract introduces a new regime called Time-Series Interaction, enabling models to continuously perceive incoming time-series data and user intent, decide when to respond, and keep processing new observations during response generation. The system employs a dual-view streaming encoder, a response control mechanism, and a decoupled inference pipeline to avoid blocking. Evaluated on the newly created StreamTSI-34K dataset, TimeInteract outperforms existing LLMs, VLMs, and TSLMs across four interaction levels, achieving significant gains in accuracy, response triggering, and inference speed.