DeepSeek code leak reveals cyber model gray-scale
Developers found hidden application entry for the 'Cyber' security model in DeepSeek's website front-end code, available in enterprise and personal versions. It is currently in gray-scale testing, requiring real-name authentication and approval to create a Cyber Key.
Summarized by AI from one outlet
Source
- IT之家Read original ↗
Latest
- Models
Researchers Simulate Nintendo Console Using Qwen Weights
Developers successfully ran and simulated a Nintendo gaming environment using approximately 1GB of Qwen model weights. This experiment demonstrates the potential of large language models for specific tasks and the viability of smaller parameter models for edge computing or novel applications.
- Models
Altman Says OpenAI Math AI Won't Stop
OpenAI CEO Sam Altman said he sympathizes with mathematicians struggling with massive AI-generated results but stated OpenAI will not stop its math AI efforts. He hinted that more discoveries are coming, indicating continued advancement of AI in mathematics.
- Models
Zhiyuan releases dual models topping multiple benchmarks
Zhiyuan (AgiBot) has launched a new dual-model system, which reportedly topped multiple public leaderboards according to Quantum Bit. This move highlights the company's progress in large language model performance, aiming to strengthen its AI capabilities.
- Models
Phonic releases voice model Phonic-1, enabling real-time search during conversation
Voice AI company Phonic unveiled its new model 'Phonic-1', which is capable of initiating searches or connecting to external services before the user finishes speaking. This feature significantly reduces the wait time for responses compared to traditional voice AI systems, improving interaction efficiency.
All news → · All rights belong to the original creators