Glock 2 Controversy Over Fake Images
Opposite Moves to Google’s Enhanced Safety Measures
Which Strategy Will Capture Market 'Interest'?
xAI, founded by Elon Musk, the founder of Tesla, has thrown down the gauntlet in the image generation artificial intelligence (AI) market. It released a trial version of its AI chatbot 'Grok2' with added image generation capabilities. The image generation model is from the German startup Black Forest Labs. It produces high-quality images comparable to Midjourney, a leading company in this field. However, it is stirring controversy due to the lack of significant restrictions on generating potentially problematic images.
According to industry sources on the 20th, fake images created with Grok2 can be found on social networking services (SNS), including former U.S. President Donald Trump wearing a Nazi uniform and singer Taylor Swift in only underwear. Images of copyrighted characters, such as Mickey Mouse smoking a cigarette, also appear. Although these could lead to defamation or portrait rights violations, xAI has not taken any particular measures.
This is gaining traction as a form of 'noise marketing.' Images generated by Grok2 are spreading across SNS and online communities. On the 15th, Musk fueled the controversy by posting on 'X' that "Grok is the most fun AI in the world." When Musk first launched Grok last year, he also highlighted its "humorous responses" as a distinguishing feature. As a latecomer, it is evaluated that he aimed for publicity.
This approach is also notable for being the exact opposite of Google's strategy. Google recently launched 'Imagen 3,' an AI tool that converts text into images. It expresses much finer details and a wider variety of styles than previous models. It can render complex textures such as fine wrinkles on human hands or knitted elephant dolls.
Unlike xAI, Google has significantly strengthened safety measures. Google DeepMind, which developed Imagen 3, devoted most of its 32-page technical report to safety. According to the report, Google filtered the training data at multiple stages to prevent the creation of harmful images. Violent or sexually explicit images were removed. To avoid biased images, AI-generated or duplicate images were also filtered out. Attention was paid to the caption texts describing the images. For example, while the text 'pig image' itself is not problematic, if the expression is combined with images of a specific race, it could lead to harmful results, so the text paired with the image was carefully checked. A separate red team was also formed to verify Imagen 3's safety. Google plans to integrate Imagen 3 into its large language model (LLM) Gemini and productivity software workspace in the future.
Google's focus on safety stems from its experience with image errors. In February, Gemini generated distorted images such as a black-skinned pope and an Indian wearing a World War II German uniform, sparking controversy. This damaged AI reliability and caused the parent company Alphabet's stock price to drop. Ultimately, Google blocked the image generation feature in Gemini. It appears that Google has strengthened safety measures to prevent such issues from recurring.
The industry is paying close attention to the outcomes of the two companies' completely opposite strategies. Image generation AI has recently been regarded as a key to monetization for big tech companies. Image generation is widely used not only in entertainment industries such as movies and games but also in corporate advertising and marketing. Unlike text generation models, where it is difficult to differentiate pricing based on response speed or length, image generation generally faces less resistance to charging fees. Grok2 is available only to subscribers of X Premium or higher, costing $7 per month. OpenAI's image generation model DALL·E 3 offers up to two free uses per day in the free version, with additional uses available in the paid version.
Jeon Chan-seok, co-CEO of Pion Corporation, said, "Although there are differences by domain, the market has grown to the extent that most places dealing with images are either considering or have already adopted generative AI."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
