A new cybersecurity report is challenging common safety beliefs, stating that even cautious smartphone owners remain vulnerable. The French government’s digital security team, CERT-FR, warns that constant Wi-Fi and Bluetooth use creates a significant, often overlooked, entry point for attackers. The authority’s latest analysis , reviewing threats since 2025, identifies public networks as especially dangerous. Hackers can deploy “man-in-the-middle” attacks on these connections to intercept private data like passwords and bank details without a user’s knowledge. A tactic called the “ Evil Twin ” attack adds to the risk. Criminals create convincing copies of legitimate hotspot names. Once a user connects to this fake network, their device can be infected with harmful software, potentially leading to a total lockdown. Connecting through older network technology, such as 2G, also increases your exposure. These outdated systems contain well-known security weaknesses that are simpler for ...
A new study reveals that artificial intelligence image generators often lack originality, repeatedly producing pictures from just a dozen common visual styles. This happens even though the technology is trained on vast libraries of online images.
Researchers played a “visual telephone” game with AI models to test their creativity. They used Stable Diffusion XL to make an image from a text prompt. Another model, LLaVA, then described that image. That description was used to create a new image, and the cycle continued for 100 rounds.
Just like the whispered message in the children’s game, the original idea quickly vanished. However, the images did not become random. Instead, they almost always shifted toward one of twelve dominant visual themes.
- PH businesses reap Php10 for every Php1 spent on AI investment — study
- AI can predict heart attack survival and personalize treatment — study
The scientists ran this experiment 1,000 separate times. In most cases, by around the 100th exchange, the output settled into a predictable style. Common results included lighthouses, fancy room interiors, cityscapes at night, and rustic buildings.
The study’s authors compared these ubiquitous styles to “visual elevator music.” They are the inoffensive, generic types of pictures often seen decorating hotel walls or corporate offices.
This trend persisted even when the researchers used different AI models for the task. It suggests a deep-seated limitation in how these systems process and recreate visual information from their training data.
The core issue may be that true creativity involves subjective taste. While AI can mimic and blend existing styles, it ultimately falls back on the most common patterns it has learned, struggling to invent genuinely new visual concepts.
Via: Gizmodo
This article, AI image generators repeatedly use the same 12 visual styles — study, was originally published at NoypiGeeks | Philippines Technology News, Reviews and How to's.
Comments
Post a Comment