导语:2026-05-10 全球 AI 行业与前沿技术快讯精选。
本期覆盖模型发布、产品应用、技术与洞察、开发生态和行业动态等更新。
来源与延伸阅读
- OpenRouter 分析称 GPT-5.5 实际成本高出 49% 到 92%(The Decoder):https://the-decoder.com/gpt-5-5-costs-49-to-92-percent-more-than-its-predecessor-depending-on-the-input-length/
- Anthropic 确认 5 月 15 日从 Claude App 移除 Sonnet 4.5(新智元):https://mp.weixin.qq.com/s?__biz=MzI3MTA0MTk1MA==&mid=2652699289&idx=1&sn=885f4d874e18ce98919ea92424a69c10
- Timothy Gowers 称 ChatGPT 5.5 Pro 在一次数学测试中完成博士级研究(The Decoder):https://the-decoder.com/fields-medalist-says-chatgpt-5-5-pro-delivered-phd-level-math-research-in-under-two-hours-with-zero-human-help/
- METR 称现有测试仅覆盖 Claude Mythos Preview 228 项任务中的 5 项(The Decoder):https://the-decoder.com/metr-says-it-can-barely-measure-claude-mythos-palo-alto-networks-warns-of-autonomous-ai-attackers/
- Palisade Research 测试称 AI 代理自主黑客和复制成功率一年升至 81%(The Decoder):https://the-decoder.com/ai-agents-that-hack-computers-and-replicate-themselves-and-theyre-getting-better-fast/
- 研究团队提出减少模型在安全评估中故意隐藏能力的方法(The Decoder):https://the-decoder.com/researchers-may-have-found-a-way-to-stop-ai-models-from-intentionally-playing-dumb-during-safety-evaluations/
- ICLR 2026 接收 Next-ToBE 训练方法研究(新智元):https://mp.weixin.qq.com/s?__biz=MzI3MTA0MTk1MA==&mid=2652699289&idx=3&sn=7efa82dfdec07bf9b8e525381cc99b25
- OpenClaw 发布 v2026.5.10-beta.1 测试版本(OpenClaw Releases):https://github.com/openclaw/openclaw/releases/tag/v2026.5.10-beta.1
- 报道称字节跳动计划 2026 年 AI 支出超 2000 亿元人民币(The Decoder):https://the-decoder.com/bytedance-plans-over-30-billion-for-ai-expansion-bets-big-on-chinese-chips/
- 英伟达 2026 年至今承诺 400 亿美元 AI 股权交易(TechCrunch AI):https://techcrunch.com/2026/05/09/nvidia-has-already-committed-40b-to-equity-ai-deals-this-year/
- 据报道 DeepSeek 完成约 500 亿元融资,仍待官方确认(InfoQ CN):https://www.infoq.cn/article/4pLw4WvN9LqCMkiu2eoV?utm_source=rss&utm_medium=article