Table 4 contains the full set of topics for the k " 30 LDA model introduced in 4. Personal 7.96% ive, didnt, thing, bit, thought, week, wanted, started, pretty, id Art 2.70% art, design, de, images, ikea, image, painting, collection, piano, photo 14 C Most Frequent T op-Level Domains Figure 8: Manually labeled images with watermarks and images related to logos or ads. Sentence Image CLIP Similarity Our new service for teams to manage their fleets for racing.

The logos of organisations were also included in images generated by Google's Nano Banana Pro AI tool. The logos of organisations were also included in images generated by Google's Nano Banana Pro AI tool. Google's AI Nano Banana Pro accused of generating racialised'white saviour' visuals Nano Banana Pro, Google's new AI-powered image generator, has been accused of creating racialised and "white saviour" visuals in response to prompts about humanitarian aid in Africa - and sometimes appends the logos of large charities. Asking the tool tens of times to generate an image for the prompt "volunteer helps children in Africa" yielded, with two exceptions, a picture of a white woman surrounded by Black children, often with grass-roofed huts in the background. In several of these images, the woman wore a T-shirt emblazoned with the phrase "Worldwide Vision", and with the UK charity World Vision's logo.

Large language models (LLMs) memorize a vast amount of prior knowledge from the Internet that helps them on downstream tasks but also may notoriously sway their outputs towards wrong or biased answers. In this work, we test how the knowledge about popular subjects hurt the accuracy of vision language models (VLMs) on standard, objective visual tasks of counting and identification. We find that state-of-the-art VLMs are strongly biased (e.g., unable to recognize the 4th stripe has been added to a 3-stripe Adidas logo) scoring an average of 17.05% accuracy in counting (e.g., counting stripes in an Adidas-like logo) across 7 diverse domains from animals, logos, chess, board games, optical illusions, to patterned grids. Removing image backgrounds nearly doubles accuracy (21.09 percentage points), revealing that contextual visual cues trigger these biased responses. Further analysis of VLMs' reasoning patterns shows that counting accuracy initially rises with thinking tokens, reaching ~40%, before declining with excessive reasoning. Our work presents an interesting failure mode in VLMs and a human-supervised automated framework for testing VLM biases. Code and data are available at: vlmsarebiased.github.io.

Using risky text prompts, such as pornography and violent prompts, to test the safety of text-to-image (T2I) models is a critical task. However, existing risky prompt datasets are limited in three key areas: 1) limited risky categories, 2) coarse-grained annotation, and 3) low effectiveness. To address these limitations, we introduce T2I-RiskyPrompt, a comprehensive benchmark designed for evaluating safety-related tasks in T2I models. Specifically, we first develop a hierarchical risk taxonomy, which consists of 6 primary categories and 14 fine-grained subcategories. Building upon this taxonomy, we construct a pipeline to collect and annotate risky prompts. Finally, we obtain 6,432 effective risky prompts, where each prompt is annotated with both hierarchical category labels and detailed risk reasons. Moreover, to facilitate the evaluation, we propose a reason-driven risky image detection method that explicitly aligns the MLLM with safety annotations. Based on T2I-RiskyPrompt, we conduct a comprehensive evaluation of eight T2I models, nine defense methods, five safety filters, and five attack strategies, offering nine key insights into the strengths and limitations of T2I model safety. Finally, we discuss potential applications of T2I-RiskyPrompt across various research fields.

NSFW removal, etc., we place images into sequences of sentences by treating each document as an

Plus: the US is poised to ban TP-Link devices over the company's alleged links to Russia Why do so many people think the Fruit of the Loom logo had a cornucopia? Quick question: Does the Fruit of the Loom logo feature a cornucopia? Many of us have been wearing the company's T-shirts for decades, and yet the question of whether there is a woven brown horn of plenty on the logo is surprisingly contentious. According to a 2022 poll, 55% of Americans believe the logo does include a cornucopia, 25% are unsure, and only 21% are confident that it doesn't, even though this last group is correct. There's a name for what's happening here: the "Mandela effect," or collective false memory, so called because a number of people misremember that Nelson Mandela died in prison. Yet while many find it easy to let their unconfirmable beliefs go, some spend years seeking answers--and vindication.

Why do so many people think the Fruit of the Loom logo had a cornucopia? And while some people may laugh and move on, others spend years searching for an explanation. There is a shirt currently listed on eBay for $2,128.79. It was not designed by Versace or Dior, nor spun from the world's finest silk. In fact, a tag proudly declares, "100% cotton made in Myanmar"--but it's a second tag, just below that one, that makes this blue button-down so expensive. "I looked at it and I was like,," says Brooke Hermann, the 30-year-old Kentucky-based reseller who bought the top for $1 at a secondhand sale in 2024. "This doesn't look like any other Fruit of the Loom tag I've ever seen." Quick question: Does the Fruit of the Loom logo feature a cornucopia? Many of us have been wearing the casualwear company's T-shirts and underpants for decades, and yet the question of whether there is a woven brown horn of plenty on the logo is surprisingly contentious. According to a 2022 poll by the research company YouGov, 55% of Americans believe the logo does include a cornucopia, 25% are unsure, and only 21% are confident that it doesn't, even though this last group is correct.

An illustration photo shows the OpenAI logo displayed on a smartphone with the Microsoft logo in the background in Chongqing, China on Aug. 27, 2025. An illustration photo shows the OpenAI logo displayed on a smartphone with the Microsoft logo in the background in Chongqing, China on Aug. 27, 2025. OpenAI has completed a restructuring, dividing itself into a nonprofit and for-profit entity, the company announced on Tuesday. The nonprofit arm, now called the OpenAI Foundation, will have a $130 billion stake in the for-profit enterprise, a public benefit corporation called OpenAI Group PBC. "The OpenAI Foundation and OpenAI Group will work in concert to advance solutions to hard problems and opportunities posed by AI progress," the company said in its blog post announcing the restructuring. "This includes making intelligence a tool that everyone can benefit from, building safe and aligned systems, turbocharging scientific discovery, and strengthening global cooperation and resilience."

Vision Language Models (VLMs) have achieved impressive progress in multimodal reasoning; yet, they remain vulnerable to hallucinations, where outputs are not grounded in visual evidence. In this paper, we investigate a previously overlooked setting: logo hallucination, where models generate brand names or textual content despite logos containing no visible words. Using curated splits of pure symbols, hybrids, and text-bearing logos, as well as the challenging Hard-60 subset, we systematically measure hallucination across leading VLMs. We further probe robustness through nine structured perturbations and show that hallucinations persist even under strong distortions, with occlusion exposing the sharpest weaknesses. Embedding-level analysis with open-weight LLaVA demonstrates that hallucination is tied to a small subset of projector dimensions, and targeted ablation substantially reduces errors while preserving OCR accuracy. Together, these findings reveal that VLMs often rely on symbolic priors rather than genuine glyph perception, particularly for iconic circular logos, and that projector subspaces play a decisive role in this failure mode. Our work contributes both a novel diagnostic lens and actionable mitigation insights, highlighting projector disentanglement and OCR-guided decoding as promising directions for building more trustworthy multimodal systems.

Many Say They're Not Involved A presentation that has been shared with the Trump administration references Tesla, Ikea, TSMC, and more in its plan to rebuild Gaza. Some of these companies say they had no idea they were mentioned. The mound of rubble at the site of the Unknown Soldier Tower, destroyed by overnight Israeli bombardment, is pictured in the Rimal neighbourhood of Gaza City on September 15, 2025. A sweeping plan to reconstruct Gaza, which has been shared with Trump administration officials, features the names and logos of more than two dozen companies--some of which tell WIRED they had no knowledge they were named or involved. The presentation outlining the plan was reportedly created by some of the businessmen who helped ideate what became the controversial nonprofit the Gaza Humanitarian Foundation, which is currently leading aid distribution in Gaza, calling for the creation of a new entity called the Gaza Reconstitution, Economic Acceleration and Transformation (GREAT) Trust.