Reality Defender CTO Alex Lisle said AI-generated food menus often look eerily flawless, with every ice cream scoop perfectly round and shrimp appearing to have been genetically modified. This unnatural aesthetic has led to a backlash in the restaurant industry, where customers find the menus unappetizing.
Lisle explained that large language models and diffusion models are trained on vast datasets, often drawing from existing menus that share a similar style. This leads to AI outputs that mimic the style of popular fast food chains, resulting in a homogenized appearance.
"It’s almost like an alien trying to make a pizza without understanding its core principles," Lisle said. The models risk model collapse when they train on too much of their own AI-generated content, leading to a degradation in quality without full failure.
Labtec demonstrated how repeated edits to AI-generated menus can make food images increasingly smooth and less realistic. The user edited a menu 100 times, resulting in images that felt unsettling and unappetizing. "The end result actually makes me uncomfortable," Labtec wrote.
Researchers at the University of Duisburg-Essen found that AI-generated food images often trigger an 'uncanny valley' effect, where almost-real images elicit more disgust than obviously fake ones. This reaction is amplified by cultural context around AI, making the issue more pronounced.
Lisle noted that the shift in how we perceive evidence, from taped confessions to AI-generated content, has fundamentally changed our trust in visual and auditory information. This has implications beyond the restaurant industry, affecting how we interpret reality.
Source: techcrunch
