推理预填充实验显示 Qwen3.8 与 GPT-5.5 Pro 重合度提升 18.18 个百分点

论文提出神经网络向量表示隐式实现了 Tensor Product Representation 符号结构，作者用 DISCOVER 方法将多种网络（含 Gemma-3-27b、GPT-2-XL、GPT-OSS-20b、Pythia-12b、Qwen3-14b、OLMo-2-13B、Llama-3.1-8b 共 7 个 LLM）的表示替换为符号化闭式方程，模型行为基本不变。

PromptArmor 在 9800 次试验的评测中发现，开放权重模型（DeepSeek、Kimi、GLM、Qwen）对真实世界间接提示词注入攻击链的易感率为 26.7%，闭源模型（GPT、Claude、Gemini）为 14.2%，开放权重模型易感率高 88%。

《经济学人》以世界价值观调查评测 25 个前沿 AI 模型。实验室来源对世界观的预测力弱于训练与对齐选择：Gemini 与 Qwen 立场接近，GPT-4o 与 DeepSeek R1 近乎相同，而 DeepSeek R1 与 DeepSeek V4 Flash 则截然不同。模型的世界观在代码生成中不可见，但在商业分析、预测、招聘与政策工作中是活跃的输入变量。

DFlash 由 UC San Diego 团队提出，是一种用于投机解码的轻量块扩散草稿模型。它一次前向推理生成整块 token，再由目标模型并行验证，保证输出无损。相比 EAGLE-3，DFlash 实现最高 2.5 倍加速，在 Qwen3-8B 等多种模型上平均无损加速超过 6 倍（MATH-500 达 6.08×）。在 NVIDIA Blackwell 上（TensorRT-LLM），gpt-oss-120b 模型吞吐量提升最高 15 倍，约为 EAGLE-3 的 1…

由Artificial Analysis和IBM推出的ITBench-AA SRE基准测试显示，所有前沿大模型得分均未超过50%。Claude Opus 4.7（自适应推理，最大努力）以47%领先，GPT-5.5（xhigh）和Qwen3.7 Max分别得46%和42%。该测试包含59个需要通过Shell命令调查Kubernetes事件快照并提交根因诊断的智能体任务。关键发现是模型推理轮次差异近3倍，但更长的轨迹并不转化为更高准确率，过度调查的模型会因提交误报而受罚。在成本…

Many recent open-weight language models use n-gram embedding tables to improve performance. These tables store learned representations of short token sequences, such as pairs or triples of consecutive tokens, which the model retrieves duri…

Computer-use agents are limited by the interfaces humans give them. An agent may type a spreadsheet cell by cell when a single line of code would do, or print all two thousand lines of a file through a shell to find one function. Both the …

Multimodal large language models (MLLMs) have made rapid progress on visual reasoning, yet many tasks require evidence that is difficult to obtain through internal reasoning alone: our controlled study on 22 benchmarks shows that simply in…

Agent learning is shifting from learning from human data toward learning through interaction with the world. However, scaling such learning directly in real-world computer systems is challenging and can introduce substantial security risks…

Large language models (LLMs) are increasingly used to automate AI research, including language-model (LM) development, offering a potential path toward recursive self-improvement. However, turning experimental feedback into a learned abili…