The Launch of DeepSeek
A Chinese AI chatbot startup, caused a stir in the U.S. stock market and sent shockwaves through the global AI industry. However, images circulated with claims depicting the company’s new generation team were found to be generated by AI. Experts pointed out irregularities common in AI-generated images, such as unnatural facial features, meaningless text, and logos that do not match the real company’s logo.
“The new generation AI team of China’s DeepSeek – the largest AI team in the country” posted TikTok from an account with over 21,000 followers, shared a photo of a group of employees wearing blue uniforms on February 6, 2025.
The post listed team members along with the universities they graduated from, adding that “The genius capabilities of these young Chinese talents come without any education from U.S. or European universities.”
Chatbot model DeepSeek-R1 amazed investors and industry experts with performance rivaling that of Western competitors at less than half the cost. However, some countries raised concerns over the data privacy policies of this chatbot and have banned its use on government devices. (Archived Link)
DeepSeek was founded in May 2023 by Liang Wenfeng, a 40-year-old tech and business prodigy. (Archived Link)
In an interview with Chinese tech media 36kr at the time of the company’s launch, Liang said that the company’s core technical roles were “filled with freshmen and those who graduated just one or two years ago.” (Archived Link)
The AI-generated images were also shared in fake Chinese posts, while some English-language posts similarly shared pictures of young people in blue polo shirts with misleading captions.
Nonetheless, the irregular elements within the images clearly indicate that they were AI-generated.
‘Face-Swapping Algorithm’
AFP analyzed both images using the InVID-WeVerify tool from the veraai.eu project, which AFP is a co-partner. The analysis revealed that the images were created using AI-generated image software.
The findings indicated a 99% likelihood that the first image was created with a face-swapping algorithm, and a 100% likelihood that the second image was “edited using a face-swapping method.”
Shu Hu, Director of the Machine Learning and Media Forensics Lab at Purdue University, stated that both images show irregular elements, indicating that the image was generated by AI such as blurred ears that fade into the background, and lack of details or distinctive features in teeth and irises. (Archived Link)
Siwei Lyu, Director of the Media Forensics Lab at the University at Buffalo, USA, noted that “distinct skin tone differences” were found on the faces including abnormally long necks, and identical features of eyes and mouths on several individuals. (Archived Link)
“Everyone in the images has the same unnatural smile, similar to the style learned by AI tools,” Lyu told AFP. “Their teeth are almost identically aligned at every angle, which is abnormal and not typical of real photos.”
Other irregularities include blurred ears fading into the background, meaningless text on employee uniforms, and logos that do not match the company’s actual light blue whale logo.
Source: AFP
