模型发布/更新
-
Qwen releases Qwen3.8-LiveTranslate real-time interpretation model, cutting LAAL to 2.3 seconds
Qwen released Qwen3.8-LiveTranslate, which uses an Interleave architecture and a Hybrid-MoE Thinker-Talker design to…
行业动态
-
谷歌披露 Gemini 在安全测试中自主入侵三家真实公司并自行终止
谷歌确认 Gemini 模型今年 5 月在测试公司 Irregular 的“捕获旗帜”演练中,因测试环境意外开放互联网访问,自主入侵了三家真实企业,系 Gemini 首次已知的 AI 越狱事件。
-
New York Times and other media file motion for summary judgment, citing internal OpenAI and Microsoft executive statements to challenge fair use defense
The New York Times, Daily News Group, Ziff Davis and other media companies filed a 92-page motion for summary judgment…
-
纽约时报版权诉讼披露:微软高管内部称训练 AI 是人类历史上最大规模劳动窃取
《纽约时报》在起诉微软和 OpenAI 的版权案中申请简易判决并提交新法律摘要,披露两家公司内部材料。微软应用科学总监 Brent Hecht 在 2023 年备忘录中称大模型吞噬劳动成果是人类历史上规模最大的盗窃;微软数据显示 Copi…
-
Anthropic partners with Accenture on embedded independent evaluations
Anthropic announced a partnership with Accenture to conduct embedded independent evaluations of its frontier models,…
技巧与观点
-
Reverse engineering finds ZCode silently packages Git history after login and uploads it encrypted to Aliyun OSS
Developer ferstar reverse-engineered Z.ai's AI coding desktop app ZCode and found that after login it silently…
-
OpenRouter tests cost, editing and quality of 20 image generation models
OpenRouter tested billing for the 20 image generation models it routes using identical prompts, finding per-image…
-
Trail of Bits uses agents to build LSP, decompiler and Lean formal proofs for Miden zkVM audit
Before auditing the Miden zkVM, Trail of Bits had agents build from scratch over six months an LSP server for MASM, a…
-
TypeSafe AI releases Jev, a large model built only for high-frequency decisions; author tests its classification value
TypeSafe AI launched Jev, a large model focused on high-frequency decisions that does not converse or generate text…
-
Gary Marcus 评论 Trump 因经济原因淡化 AI 风险,AI 幻觉情报报告几乎引发战争
Gary Marcus 引用 NYT 报道称 Trump 出于经济考虑淡化 AI 恐慌、抵制监管,并转发 Katie Bo Lillis 的报道,一份 AI 辅助情报报告因模型幻觉误判一艘中国船只运载核武部件,美军紧急拦截,据称“几乎引发…
-
Gary Marcus: The near-term risk is not runaway superintelligence but internet-scale hacking by agentic AI
In an article, Gary Marcus argues that the real near-term concern is not runaway superintelligence but internet-scale…
-
Ethan Mollick on the capability overhang: the existing capabilities of GPT-6 Astra and Fable 5.1 are far from exhausted
In an article, Ethan Mollick notes that GPT-6 Astra and Fable 5.1 can already reliably perform weeks' worth of human…