Archive
GitHub 熱門專案、開源工具、框架、runtime 與重要 release。
最新模型
RuleMaze uses executable verifiers to test whether multimodal models can understand mazes, follow natural-language rules, and complete path planning. By separating perception, movement, and rule verification, DMP substantially outperforms standard fine-tuning, though testing remains limited to synthetic grids and the verifiers underwent manual correction.
代理工具
AI4AI-Bench evaluates hyperparameter tuning separately from genuine changes to learning methods, requiring coding agents to submit training code that can be rerun from scratch. Across 290 trials involving six systems, more than 40% underperformed the original algorithms, while increased reasoning effort mainly made agents more willing to modify core training mechanisms.
代理工具
Brain Researcher constrains neuroimaging analysis agents with a knowledge graph, typed workflows, and review rules, while preserving hypotheses, evidence, and execution provenance. Tool routing improved substantially, but the rate of successfully grounding claims in verifiable evidence remained just 22.0%, and automated review also failed to catch a directional statistical error.
AI 安全
SafeBranch rolls back an agent’s hazardous trajectory to a critical decision point and compares safe and unsafe actions only within the same context. The trained model requires no external critic at deployment, but results are currently limited to simulated household environments, and data construction still relies on GPT-4o judgments.
推論系統
FormalTCS extracts research-grade problems from top theoretical computer science papers published over the past two years, measuring models stage by stage as they turn natural-language claims into machine-verifiable proofs. The best model achieves a Pass@8 of 28.6 when proving already-formalized theorems, but translating natural language into correct Lean theorem statements remains the primary bottleneck.
AI 安全
The open-source agent runtime Forge can now isolate MCP connections and credentials by user while automatically performing OAuth 2.1 metadata discovery and dynamic client registration. The release also adds granular egress controls and approval auditing, although MCP support currently remains focused on HTTP transport.
AI 開發者平台
The new release raises the minimum Python version to 3.10 and removes Text Completions, several sampling parameters, and the client-side compaction interface. Most standard API calls require no changes, but monitoring, testing, and custom transport layers that depend directly on httpx must be migrated.
推論系統
Daedalus-150M is designed specifically for single-user, 4-bit CPU inference, allowing most network layers to avoid repeatedly reading an ever-growing KV cache. Preliminary results indicate that its speed advantage increases with context length, but the model is limited to 2K context, and its performance figures still await independent reproduction across different hardware.
AI 安全
The new benchmark deduplicates agent skills from 13 public sources and splits them by structural family, preventing near-duplicate samples from appearing in both the training and test sets. The best text classifier performed strongly on a random split, but misclassified 62.4% of benign skills as malicious when evaluated on unseen sources.
代理工具
NVIDIA has updated its open-source agent execution stack, focusing on state drift across vLLM, llama.cpp, Ollama, and the model router. The new release also tightens sandbox lifecycle management, image provenance, and failure recovery, though it remains a patch release in a rapidly evolving project.
代理工具
SWE-bench Science uses 98 real-world scientific software repositories to test implicit contracts involving units, coordinate systems, numerical invariants, file formats, and more. Claude Code with Opus 5 currently ranks first, but its pass@1 remains below 50%, falling to just 27.78% on engineering integration tasks.
代理工具
Microsoft has open-sourced Thinkingbox, which uses isolated MCP tool environments and executable assertions to verify whether agents actually complete their work, rather than merely checking their responses or whether tool calls are valid. The best system achieves a 65.36% single-run success rate, but that falls to just 25.25% when the same task must succeed 20 consecutive times.