The American Writers Association Launches a Survey on Generative AI to Protect Writers' Rights and Literary Diversity


iFlytek released Spark-Audio-1.0-Preview, a speech foundation model trained on domestic computing power to reduce supply chain risks. It supports text/speech input and text output, with speech transcription, multilingual translation, dialect recognition, environmental sound recognition, speaker recognition, sentiment analysis, and audio Q&A.....
iFlytek has launched the speech base large model Spark-Audio-1.0-Preview, trained using entirely domestic computing power. Traditional audio processing often adopts a cascading approach, first converting speech to text and then understanding it, which has two major shortcomings: information loss (ignoring tone, emotion, background sounds) and fragmented processes (separate stages such as speech transcription and voiceprint recognition), affecting the completeness of scenario judgment.
On Sept 16, Alibaba's Qwen3.8-27B topped Hugging Face's most popular open-source model list, surpassing FLUX.1, DeepSeek-R1, Kimi-K3, and Meta-Llama-3. Ranked by developer likes, it reflects real preference over benchmarks or parameters, becoming the community's most popular open-source model ever.....
Security workers serving California offices of OpenAI, Google, Anthropic and others plan to strike this week, filing 72-hour notice on the 15th and refusing shifts from 4:30 a.m. on the 18th, per Business Insider. SEIU-USWW says Amazon, Microsoft, Salesforce, Meta, Nvidia are also affected.....
Anthropic signed its first Australian data center lease, securing 2.16 GW capacity, with operations planned for 2027 on farmland about 250 km from Brisbane. Following OpenAI, it expands in Oceania, drawn by power and land, to diversify training and inference compute.....