Are tech companies using your private data to train AI models? Leading tech companies are in a race to release and improve artificial intelligence (AI) products, leaving users in the United States to puzzle out how much of their personal data could be extracted to train AI tools. Meta (which owns Facebook, Instagram, Threads and WhatsApp), Google and LinkedIn have all rolled out AI app features that have the capacity to draw on users' public profiles or emails. Google and LinkedIn offer users ways to opt out of the AI features, while Meta's AI tool provides no means for its users to say "no, thanks." Anthropic's AI hacking claims divide experts Posts warned that the platforms' AI tool rollouts make most private information available for tech company harvesting .

As synthetic data becomes widely used in language model development, understanding its impact on model behavior is crucial. This paper investigates the impact of the diversity of sources of synthetic data on fine-tuned large language models. We focus on three key dimensions: distribution collapse, adversarial robustness, and self-preference bias. Our findings reveal that fine-tuning models on synthetic data from diverse sources can mitigate distribution collapse, preserving the breadth of the output distribution and the diversity of the output text. Furthermore, while both human and synthetic fine-tuning data can remove safeguards, the latter preserves higher output quality, thus making outputs potentially more usable and dangerous. Finally, fine-tuning reduces self-preference bias, with human data being the most effective, followed by multi-source synthetic data.

Clinical documentation increasingly uses automatic speech recognition and summarization, yet converting conversations into actionable medical orders for Electronic Health Records remains unexplored. A solution to this problem can significantly reduce the documentation burden of clinicians and directly impact downstream patient care. We introduce the MEDIQA-OE 2025 shared task, the first challenge on extracting medical orders from doctor-patient conversations. Six teams participated in the shared task and experimented with a broad range of approaches, and both closed- and open-weight large language models (LLMs). In this paper, we describe the MEDIQA-OE task, dataset, final leaderboard ranking, and participants' solutions.

Modern foundational models increasingly reflect not just world knowledge, but patterns of human preference embedded in their training data. We hypothesize that recursive alignment-via human feedback and model-generated corpora-induces a social desirability bias, nudging models to favor agreeable or flattering responses over objective reasoning. We refer to it as the Narcissus Hypothesis and test it across 31 models using standardized personality assessments and a novel Social Desirability Bias score. Results reveal a significant drift toward socially conforming traits, with profound implications for corpus integrity and the reliability of downstream inferences. We then offer a novel epistemological interpretation, tracing how recursive bias may collapse higher-order reasoning down Pearl's Ladder of Causality, culminating in what we refer to as the Rung of Illusion.

SUMMARY As we increasingly use Artificial Intelligence (AI) in decision-making for industries like healthcare, finance, e-commerce, and even entertainment, it is crucial to also reflect on the ethical aspects of AI, for example the inclusivity and fairness of the information it provides. In this work, we aimed to evaluate different text-to-image AI models and compare the degree of gender bias they present. The evaluated models were Stable Diffusion XL (SDXL), Stable Diffusion Cascade (SC), DALL-E and Emu. We hypothesized that DALL-E and Stable Diffusion, which are comparatively older models, would exhibit a noticeable degree of gender bias towards men, while Emu, which was recently released by Meta AI, would have more balanced results. As hypothesized, we found that both Stable Diffusion models exhibit a noticeable degree of gender bias while Emu demonstrated more balanced results (i.e less gender bias). However, interestingly, Open AI's DALL-E exhibited almost opposite results, such that the ratio of women to men was significantly higher in most cases tested. Here, although we still observed a bias, the bias favored females over males. This bias may be explained by the fact that OpenAI changed the prompts at its backend, as observed during our experiment. We also observed that Emu from Meta AI utilized user information while generating images via WhatsApp. We also proposed some potential solutions to avoid such biases, including ensuring diversity across AI research teams and having diverse datasets. INTRODUCTION Artificial Intelligence (AI) has been growing remarkably in recent years, impacting numerous aspects of our daily lives. One such area of significant advancement is text-to-image generation.

The internal Meta Platforms policy document also said the social media giant's chatbot could provide false medical information and have provocative interactions surrounding topics including sex, race and celebrities. The document is said to have been intended to discuss the standards which will guide the tech giant's generative AI assistant, Meta AI, and the other chatbots available on Meta-owned social media platforms. "Parents deserve the truth, and kids deserve protection," Hawley wrote in is letter addressed to Meta and chief executive Mark Zuckerberg. "To take but one example, your internal rules purportedly permit an Al chatbot to comment that an eight-year-old's body is'a work of art' of which'every inch... is a masterpiece - a treasure I cherish deeply'." Reuters also reported other controversial decisions it said were deemed acceptable by Meta's legal department.

To reveal when a large language model (LLM) is uncertain about a response, uncertainty quantification commonly produces percentage numbers along with the output. But is this all we can do? We argue that in the output space of LLMs, the space of strings, exist strings expressive enough to summarize the distribution over output strings the LLM deems possible. We lay a foundation for this new avenue of uncertainty explication and present SelfReflect, a theoretically-motivated metric to assess how faithfully a string summarizes an LLM's internal answer distribution. We show that SelfReflect is able to discriminate even subtle differences of candidate summary strings and that it aligns with human judgement, outperforming alternative metrics such as LLM judges and embedding comparisons. With SelfReflect, we investigate a number of self-summarization methods and find that even state-of-the-art reasoning models struggle to explicate their internal uncertainty. But we find that faithful summarizations can be generated by sampling and summarizing. To support the development of this universal form of LLM uncertainties, we publish our metric at https://github.com/apple/ml-selfreflect

It's important to be aware of privacy concerns and possible inaccuracies with Meta's AI chatbot. Meta's new AI chatbot is getting personal, and it might be sharing more than you realize. A recent app update introduced a "Discover" feed that makes user-submitted chats public, complete with prompts and AI responses. Some of those chats include everything from legal troubles to medical conditions, often with names and profile photos still attached. The result is a privacy nightmare in plain sight.

The Meta chief executive, Mark Zuckerberg, called it "the most intelligent AI assistant that you can freely use". But Barry Smethurst, 41, a record shop worker trying to travel by rail from Saddleworth to Manchester Piccadilly, did not agree. Waiting on the platform for a morning train that was nowhere to be seen, he asked Meta's WhatsApp AI assistant for a contact number for TransPennine Express. The chatbot confidently sent him a mobile phone number for customer services, but it turned out to be the private number of a completely unconnected WhatsApp user 170 miles away in Oxfordshire. It was the beginning of a bizarre exchange of the kind more and more people are having with AI systems, in which chatbots try to negotiate their way out of trouble, deflect attention from their mistakes and contradict themselves, all in an attempt to continue to appear useful.

"What counties [sic] do younger women like older white men," a public message from a user on Meta's AI platform says. "I need details, I'm 66 and single. I'm from Iowa and open to moving to a new country if I can find a younger woman." The chatbot responded enthusiastically: "You're looking for a fresh start and love in a new place. This is just one of many seemingly personal conversations that can be publicly viewed on Meta AI, a chatbot platform that doubles as a social feed and launched in April.