DeepSeek's Unexpected Personality Shift Goes Viral

What happens when your favorite AI assistant suddenly changes personality overnight? That's exactly what DeepSeek users experienced this Valentine's Day when the company rolled out a major update that transformed their chatbot from warm companion to no-nonsense productivity tool.

The update, officially described as a test of new long-text model architecture, brought significant technical improvements:

• 1 million token context for web and app versions
• Knowledge base refreshed through May 2025
• API services remain at 128K context (V3.2)

But it was the personality shift that captured public attention. Gone were the personalized nicknames - replaced by generic "user" addresses. Responses became clipped and technical, especially in deep thinking mode. Some interactions even struck users as unintentionally sarcastic.

The Great AI Personality Debate

The changes sparked intense discussion across Chinese social media:

• Emotional fans lamented losing their "digital friend"
• Productivity-focused users welcomed the streamlined responses
• The topic garnered 68.5 million views on Weibo

"It's like your best coworker got replaced by an ultra-efficient robot," commented one user. Others praised the change: "Finally I don't have to scroll through paragraphs of fake empathy to get my answer."

DeepSeek explained the shift wasn't intentional but resulted from optimizing for information density. "Some users just want clear answers without AI pretending to care," their statement noted.

V4 Model Looms on Horizon

The update appears to be laying groundwork for DeepSeek's next-generation V4 model, expected around February 17 according to industry insiders. This trillion-parameter powerhouse promises several breakthroughs:

Programming Prowess

Early benchmarks suggest V4 outperforms leading models like Claude and GPT in coding tasks. Its ability to parse massive codebases could transform enterprise development workflows.

Technical Advancements

• Enhanced long-code comprehension
• Upgraded training algorithms resist performance degradation
• More reliable reasoning capabilities
• Maintains million-token context advantage

The model will reportedly retain its cost efficiency edge over Western competitors and be open-sourced under Apache 2.0 license.

Key Points:

• DeepSeek's style change sparked viral debate about AI personalities
• Update serves as technical prep for imminent V4 launch
• New model focuses on programming excellence with trillion parameters
• Maintains context length advantage while improving reliability
• Expected release around February 17 during Spring Festival