DeepSeek Responds to Model Abnormal Response: Special Characters Caused Hallucinations, No Security Risks

Recently, several users encountered a false alarm while using the web version of the popular large model DeepSeek. After entering specific special characters in the chat box, the system unexpectedly loaded completely irrelevant text.

Since these automatically popping-up responses looked very much like other users' questions, it sparked widespread concerns and heated discussions online about "dialogue privacy leakage."

Official Investigation Identifies Hallucination

Facing user concerns, the DeepSeek official team quickly launched a comprehensive investigation and released an official statement on the anomaly response early today. The official clearly stated that the abnormal content triggered by entering specific special characters is essentially a model hallucination caused by special characters.

The technical team emphasized that this incident was purely a cognitive bias of the model when processing specific symbols, and absolutely no security issues occurred, nor was there any user privacy leak.

Targeted Training to Optimize Experience

To completely resolve this issue affecting user experience, DeepSeek stated that it will enhance the model's capabilities through targeted training. The technical team will focus on optimizing the model's recognition and processing logic for special characters, and fix the known issues as soon as possible.

Notably, DeepSeek recently launched a new V4 preview version of the large model and also open-sourced it, featuring ultra-long context capability of millions of words. Finally, the official expressed gratitude to users for their supervision and promised to always place high importance on user data security and usage experience.

The Race for AI Talent Heats Up: Yue Zhi Anmian and DeepSeek's Path of Anti-Global Tech Giants

On May 12, Zhang Yutong, VP of Moonshot AI, held a recruitment event at Peking University, engaging with Guanghua School of Management Dean Tian Xuan and offering on-site interviews. This underscores Moonshot AI's accelerated push in the AI talent war, as China's large model industry booms with fierce competition among leading firms like Moonshot AI and DeepSeek driving tech-focused hiring.....

Lee Kai-Fu: Open Source Models Are a Better Path to AI Sovereignty

Lee Kai-Fu introduced the concept of "AI Sovereignty," emphasizing that it involves technological control, data security, and the adaptation of models to local cultural and legal frameworks. He believes that countries do not need to blindly reinvent OpenAI, as closed-source self-development is costly and unrealistic. For countries and companies with limited resources, building a localized system based on open source models is a more feasible third option.

Liang Wenfeng Invests 20 Billion! DeepSeek Launches Record-Breaking $5 Billion Funding Round, V4.1 Set for June

The domestic large model sector has witnessed a capital storm, with DeepSeek (Deep Seek) initiating its first major funding round targeting an amount of 5 billion yuan, which, if successful, would break the industry record. Notably, the lead investor is not a venture capital firm or an internet giant, but the founder Liang Wenfeng himself, who has invested the highest personal amount, demonstrating his strong confidence in the company.

DeepSeek Opens Wide-Ranged Image Recognition Mode: Multimodal Understanding Function Enters Internal Testing

DeepSeek launched a large-scale beta of its 'Image Recognition Mode' on May 9, marking its entry into multimodal interaction. After a limited gray test in late April, most accounts can now access it via a standalone entry in the chat interface. Though still in beta, it's listed alongside 'Fast Mode' and 'Expert Mode,' signaling multimodal understanding as a core direction.....