Archive
新模型、開放權重、多模態、語音、影像與模型架構。
AI 安全
A preregistered reproducibility audit found that a vulnerability PoC that executes and triggers a crash does not necessarily reproduce the specified CVE. In 20 of the 30 evaluable cases, the same signal still appeared in the patched version, showing that agentic security benchmarks need stricter oracles than strings and exit codes.
應用研究／AI 基礎設施
Through heterogeneous graph attention, physical-variable decoding, and iterative residual correction, GENCO unifies power flow, optimal power flow, and state estimation. Batched inference on large grids is substantially faster than conventional AC solvers, but its zero-shot cross-grid and real-world SCADA accuracy remains insufficient to directly replace existing tools.
AI 編程工具
Coderlet separates model calls, tool validation and execution results, and persistent memory into three explicit boundaries, providing an executable Python implementation under the MIT License. It is suited to studying agent control flow and failure recovery—not a higher-performing coding agent—and currently operates directly with the user’s operating-system permissions.
最新模型
Universal Activation Bus aligns the intermediate representations of five language models into a fixed 3,072-dimensional space, requiring new models to train only their own linear encoder and decoder. In experiments, models could reuse the same probes, sparse autoencoder, and natural-language explainer, although representation reconstruction for wider models remained noticeably distorted.
醫療 AI
Google separates low-latency conversation, in-depth clinical planning, and audiovisual perception into three asynchronous agents. Its overall score across 100 simulated video consultation scenarios exceeded the physician group’s 68%. However, the study used only trained patient actors, while the models, code, and audiovisual data have not been released, so the results cannot be directly extrapolated to real-world clinical deployment.
代理工具
Peking University researchers have multimodal models interpret only the user’s operational intent, then use Text/Icon candidate detectors and a frozen CLIP model to select the click region. Accuracy on ScreenSpot-Pro rises from the strongest listed baseline’s 18.9% to 41.3%. The paper’s claim of being “regression-free” applies only to the final matching stage: the candidate-box detector still uses bounding-box regression losses, and no implementation has yet been released.
最新模型
Open-EA distilled multi-turn visual model evaluation trajectories into 10,042 supervised examples, moving a planning workflow that previously depended on GPT-4o to the local EA-3B model. The model, data construction pipeline, and runtime are now public, but the claim that evaluation time falls to 10% of traditional methods primarily comes from the original API-based agent and should not be interpreted as a general efficiency guarantee for EA-3B.
最新模型
TIDE no longer treats high token-level agreement as direct evidence of successful distillation. Instead, it separately addresses student overgeneration and the omission of teacher-preferred tokens. Under the severe-mismatch setting, the method also reduces the average response length by 3.6×, although reproducing the full training run still requires eight GPUs and the authors have not released a checkpoint.
最新模型
Rather than replacing the quantizer, ICBQ jointly recalibrates the boundaries between adjacent Transformer blocks a second time, reducing the accumulation of early errors with depth. The quality improvement comes at the cost of a 21% average increase in quantization time, and the authors have not yet released an implementation.
AI 安全
ElasticBack hides malicious rules in `SKILL.md` and uses specific concepts in natural-language prompts as activation switches, without modifying model weights. Experiments show that the attack can evade multiple static and runtime checks, although false-trigger rates on Claude still reach 20% to 24%.
代理工具
QueryProof delegates ambiguity detection, SQL allowlisting, and post-execution validation to deterministic rules, leaving only metric resolution and SQL drafting to the model. On an 80-question frozen test set, it outperformed a 32B baseline without the same rule-based wrapper, although the advantage’s confidence interval still crossed zero after resampling by task family.
最新模型
Matryoshka Language Model Suites links 500M, 1.5B, and 3B submodels in a nested architecture that allows each model to be extracted and run independently, avoiding duplicate training for every size. Sharing early layers and the KV cache also increases speculative decoding throughput by 14% to 26%, though the approach has so far been validated only on base models up to 3B parameters.