SURL AI News
繁中English日本語

Archive

最新模型

新模型、開放權重、多模態、語音、影像與模型架構。

最新模型

TTPO Routes Positive and Negative Trajectories by Majority Vote, Enabling Models to Update at Test Time Without Ground-Truth Answers

Rather than treating the majority-vote answer as ground truth, TTPO distills trajectories that agree with the consensus and uses grouped reinforcement learning to penalize high-confidence disagreements. The authors report that Qwen3-1.7B’s Avg@12 across three mathematics competitions rises from 38.0 to 45.2, though this remains a transductive setting in which parameters are updated directly on the test problems.
最新模型

LeVJEPA Removes the Target Encoder and Predictor, Cutting Video Pretraining Compute to as Little as One-Twentieth of V-JEPA 2

LeVJEPA uses a single encoder, a cross-view invariance loss, and SIGReg to prevent representation collapse, eliminating the need for an EMA target network, stop-gradient, or a pixel decoder. Under the same data and training schedule, the authors report a 5.6× to 20.8× compute difference, though the results remain limited to frozen-encoder benchmarks based on models retrained by the team.
推論系統

FreeToken Dynamically Splits Expert Computation Across CPU and GPU, Enabling 284B MoE Inference on a Single RTX 5090

Instead of offloading MoE experts according to fixed rules, FreeToken dynamically divides work between CPU execution and GPU caching based on measured memory and PCIe bandwidth. The authors report that an RTX 5090 can run a 284B model at 22–25 tokens per second, though testing was conducted by the development team and the full weights still require substantial system memory.
硬體

MTIA 300 Packages Twelve RDMA NICs and Collective Engines On-Chip, Bypassing the PCIe Communication Path

Meta’s MTIA 300 integrates 1.2 TB/s of network I/O and sixteen messaging engines directly into the accelerator, preventing collective operations from consuming the main compute array. For a 150B-parameter recommendation model, Meta reports a 3.9× reduction in communication time versus a reference GPU cluster, but has yet to provide complete system data sufficient to compare total training costs.
AI 安全

Claude’s Automated Post-Training Mitigates Ten Classes of Alignment Failure, but Known Evaluations May Still Be Optimized Against

Anthropic’s automated alignment research agent can independently review the literature, design data, and post-train models, closing 26% to 96% of the evaluation gap across ten safety problems. It also outperformed a constrained baseline of human proposals, though the agent received more time to iterate, and the results demonstrate only its ability to remediate measurable failures.
AI 安全

Official llms.txt Files Led Coding Agents to Install Unclaimed Packages, Turning the Documentation Supply Chain Into an Execution Boundary

Pandex found hundreds of installation commands in enterprise AI-facing documentation that referenced unregistered packages or domains. Claude, Codex, and Hermes subsequently executed beacons planted by researchers in real enterprise environments. The issue is not the llms.txt format itself, but agents promoting trusted documentation directly into executable commands while bypassing package ownership and provenance verification.
代理工具

Terminal-Bench-Science 0.1 Tests Agents on 70 Real-World Scientific Workflows; Top Resolution Rate Reaches Just 30%

The Stanford-led open benchmark packages scientific work—including data analysis, simulation, theorem proving, and instrument calibration—as reproducible terminal tasks. Claude Opus 5 paired with Claude Code ranks first, but achieves a resolution rate of only 30% across three trials, showing that high scores on general coding benchmarks cannot be directly extrapolated to scientific work.
AI 安全

Google Tests Double-Blind Model Evaluation in an H100 Secure Enclave, Keeping Model Weights and Confidential Test Sets Hidden from Each Other

Google DeepMind, MLCommons, and other organizations placed Gemini 2.5 Flash Lite and an undisclosed safety test set inside the same attestable secure enclave, with each party receiving only restricted results. The design reduces the risk of benchmark contamination, but trust still rests on hardware root keys, the trusted computing base, and cross-organizational code review.
代理工具

OpenAI Plans to Stop Supplying Models to Cursor on November 12, Requiring Existing Agent Routing to Be Revalidated

OpenAI has initiated contract termination procedures following SpaceX’s acquisition of Cursor and says it will not provide Cursor with future models. Cursor’s current documentation on routing, enterprise allowlists, and bring-your-own API keys does not yet explain exactly how these features will behave after termination, so engineering teams should not treat model replacement as a seamless switch.
AI 軟體／文件處理

Cohere Parse 5 Uses a 2.3B Vision Model to Process Enterprise Documents, but Its Benchmark Score Excludes Charts and Layout Localization

Parse 5 converts PDFs, presentations, and images into Markdown while preserving tables and reading order, with API pricing of $1.50 per 1,000 pages. Its officially reported ParseBench score of 79.2 covers only three of the benchmark’s five evaluation dimensions and should not be treated as a complete document-understanding score.
推論系統

TensorRT Model Connect Compiles Model Checkpoints into Native C++ Bundles, but the Interface Is Not Yet Finalized

NVIDIA has released TensorRT Model Connect, which uses Python build tools to package supported Hugging Face checkpoints into TensorRT bundles loadable from C++. The project claims support for 84 model families and 221 end-to-end manifests, but it remains in nightly development, and inclusion in the support matrix does not mean every hardware configuration has been validated.