Recent advances in generative models have demonstrated remarkable capabilities in producing high-quality images, but their reliance on large-scale unlabeled data has raised significant safety and copyright concerns. Efforts to address these issues by erasing unwanted concepts have shown promise. However, many existing erasure methods involve excessive modifications that compromise the overall utility of the model. In this work, we address these issues by formulating a novel minimalist concept erasure objective based \emph{only} on the distributional distance of final generation outputs. Building on our formulation, we derive a tractable loss for differentiable optimization that leverages backpropagation through all generation steps in an end-to-end manner. We also conduct extensive analysis to show theoretical connections with other models and methods. To improve the robustness of the erasure, we incorporate neuron masking as an alternative to model fine-tuning. Empirical evaluations on state-of-the-art flow-matching models demonstrate that our method robustly erases concepts without degrading overall model performance, paving the way for safer and more responsible generative models.

Large language models (LLMs) can elicit social bias during generations, especially when inference with toxic prompts. Controlling the sensitive attributes in generation encounters challenges in data distribution, generalizability, and efficiency. Specifically, fine-tuning and retrieval demand extensive unbiased corpus, while direct prompting requires meticulously curated instructions for correcting the output in multiple rounds of thoughts but poses challenges on memory and inference latency. In this work, we propose the Expert-Guided Extinction of Toxic Tokens for Debiased Generation (EXPOSED) to eliminate the undesired harmful outputs for LLMs without the aforementioned requirements. EXPOSED constructs a debiasing expert based on the abundant toxic corpus to expose and elicit the potentially dangerous tokens. It then processes the output to the LLMs and constructs a fair distribution by suppressing and attenuating the toxic tokens. EXPOSED is evaluated on fairness benchmarks over three LLM families. Extensive experiments demonstrate that compared with other baselines, the proposed EXPOSED significantly reduces the potential social bias while balancing fairness and generation performance.

Accountants are among the professionals whose careers are most exposed to the capabilities of generative artificial intelligence, according to a new study. The researchers found that at least half of accounting tasks could be completed much faster with the technology. The same was true for mathematicians, interpreters, writers and nearly 20% of the U.S. workforce, according to the study by researchers at the University of Pennsylvania and OpenAI, the company that makes the popular AI tool ChatGPT.

Game theory has been an effective tool in the control of disease spread and in suggesting optimal policies at both individual and area levels. In this AMS Notices article, we focus on the decision-making development for the intervention of COVID-19, aiming to provide mathematical models and efficient machine learning methods, and justifications for related policies that have been implemented in the past and explain how the authorities' decisions affect their neighboring regions from a game theory viewpoint.

Cense.ai is an Artificial Intelligence company that works in a wide range of areas. According to the company website, Cense.ai It is this last practice that led to the company exposing over 2.5 million medical records. According to researcher Jeremiah Fowler, all of the records were readily available to view or download by anyone with an Internet connection. Though it remains unclear how long the data was available online, Fowler made the discovery on July 7th, 2020.

The commonly used metrics for motion prediction do not correlate well with a self-driving vehicle's system-level performance. The most common metrics are average displacement error (ADE) and final displacement error (FDE), which omit many features, making them poor self-driving performance indicators. Since high-fidelity simulations and track testing can be resource-intensive, the use of prediction metrics better correlated with full-system behavior allows for swifter iteration cycles. In this paper, we offer a conceptual framework for prediction evaluation highly specific to self-driving. We propose two complementary metrics that quantify the effects of motion prediction on safety (related to recall) and comfort (related to precision). Using a simulator, we demonstrate that our safety metric has a significantly better signal-to-noise ratio than displacement error in identifying unsafe events.

Smart-assistant devices have had their share of privacy missteps, but they're generally considered safe enough for most people. New research into vulnerabilities in Amazon's Alexa platform, though, highlights the importance of thinking about the personal data your smart assistant stores about you--and minimizing it as much as you can. Findings published on Thursday by the security firm Check Point reveal that Alexa's web services had bugs that a hacker could have exploited to grab a target's entire voice history, meaning their recorded audio interactions with Alexa. Amazon has patched the flaws, but the vulnerability could have also yielded profile information, including home address, as well as all of the "skills," or apps, the user had added for Alexa. An attacker could have even deleted an existing skill and installed a malicious one to grab more data after the initial attack.

Facebook is still a privacy nightmare. The company's history of porous data sharing continues to haunt both it and us (its fairly helpless users) on the regular. On Wednesday, researchers from the cybersecurity firm UpGuard shared that they found two massive troves of exposed Facebook user data that had been posted publicly on Amazon cloud servers. The data included users' passwords, names, comments, and likes. The scope of this particular privacy foul from Facebook is tremendous: More than 540 million user records were sitting in plain sight, available to anyone who found them.

Brash Games is no more. The video game news site that claimed to be a source for unbiased game reviews has folded its website and its domain is now for sale. Just less than 24 hours ago, Brash Games was still putting up reviews for "Star Wars: Galactic Battlegrounds Saga" and "Flinthook" on its website, but now when gamers try to access the site they end up loading a page that says the domain name for Brash Games (brashgames.co.uk) is up for grabs. The domain listing, which is courtesy of GetDotted.com, The shutdown comes after the website came under fire when its shady business and unethical practices got exposed.