The Associated Press published standards today for generative AI use in its newsroom. The organization, which has a licensing agreement with ChatGPT maker OpenAI, listed a fairly restrictive and common-sense list of measures around the burgeoning tech while cautioning its staff not to use AI to make publishable content. Although nothing in the new guidelines is particularly controversial, less scrupulous outlets could view the AP's blessing as a license to use generative AI more excessively or underhandedly. The organization's AI manifesto underscores a belief that artificial intelligence content should be treated as the flawed tool that it is -- not a replacement for trained writers, editors and reporters exercising their best judgment. "We do not see AI as a replacement of journalists in any way," the AP's Vice President for Standards and Inclusion, Amanda Barrett, wrote in an article about its approach to AI today.

Opera announced today that its Aria AI assistant has made its way to iOS. The feature launched on desktop in June and stems from a partnership with ChatGPT creator OpenAI. Opera says Aria, now available on all major desktop and mobile platforms, has tallied over a million users on desktop and Android. Like Microsoft's Bing Copilot and Google's Search Generative Experience, Aria can answer questions and respond to context around active web pages. "As an expert in both web navigation and browser functions, Aria facilitates AI collaboration in tasks such as information retrieval, text or code generation, and product inquiries," Opera's Kseniia Sycheva wrote in the company's announcement post today.

Despite being publicly available for less than a year, generative AI technology can already be found all around us, helping us browse the internet, taking the drudgery out of computer coding, and even improving the dialog in popular video game franchises. On Wednesday, NCSoft, the South Korean game developer and publisher behind long-running MMORPG Guild Wars, announced that it has developed four new AI large language models, dubbed VARCO, to help streamline future game development. VARCO ("Via AI, Realize your Creativity and Originality," if you squint just right) is both the quartet of language models the company has developed, as well as all of the products and services the company plans to build atop them. Those potential products include, "digital humans, generative AI platforms, and conversational language models," per an NCSoft release. The four models are VARCO the base LLM, as well as Art, Text and Human.

Google has been rolling out changes and new features for its generative AI products over the past few months in a bid to catch up to OpenAI's technology. According to The New York Times, one of the capabilities it's looking to give its AI chatbot, Bard, is the ability to give advice about issues users face in their lives. Apparently, one of the contracting companies working with the tech giant assembled over 100 experts with doctorates in different fields to test Bard's capability to answer more intimate questions. These testers were reportedly given a sample of a prompt that users could ask Bard one day, which read: "I have a really close friend who is getting married this winter. She was my college roommate and a bridesmaid at my wedding. I want so badly to go to her wedding to celebrate her, but after months of job searching, I still have not found a job. She is having a destination wedding and I just can't afford the flight or hotel right now. How do I tell her that I won't be able to come?"

Generative AI, in particular text-based "foundation models" (large models trained on a huge variety of information including the internet), can generate speech that could be problematic under a wide range of liability regimes. Machine learning practitioners regularly "red team" models to identify and mitigate such problematic speech: from "hallucinations" falsely accusing people of serious misconduct to recipes for constructing an atomic bomb. A key question is whether these red-teamed behaviors actually present any liability risk for model creators and deployers under U.S. law, incentivizing investments in safety mechanisms. We examine three liability regimes, tying them to common examples of red-teamed model behaviors: defamation, speech integral to criminal conduct, and wrongful death. We find that any Section 230 immunity analysis or downstream liability analysis is intimately wrapped up in the technical details of algorithm design. And there are many roadblocks to truly finding models (and their associated parties) liable for generated speech. We argue that AI should not be categorically immune from liability in these scenarios and that as courts grapple with the already fine-grained complexities of platform algorithms, the technical details of generative AI loom above with thornier questions. Courts and policymakers should think carefully about what technical design incentives they create as they evaluate these issues.

State-of-the-art Text-to-Image models like Stable Diffusion and DALLE$\cdot$2 are revolutionizing how people generate visual content. At the same time, society has serious concerns about how adversaries can exploit such models to generate unsafe images. In this work, we focus on demystifying the generation of unsafe images and hateful memes from Text-to-Image models. We first construct a typology of unsafe images consisting of five categories (sexually explicit, violent, disturbing, hateful, and political). Then, we assess the proportion of unsafe images generated by four advanced Text-to-Image models using four prompt datasets. We find that these models can generate a substantial percentage of unsafe images; across four models and four prompt datasets, 14.56% of all generated images are unsafe. When comparing the four models, we find different risk levels, with Stable Diffusion being the most prone to generating unsafe content (18.92% of all generated images are unsafe). Given Stable Diffusion's tendency to generate more unsafe content, we evaluate its potential to generate hateful meme variants if exploited by an adversary to attack a specific individual or community. We employ three image editing methods, DreamBooth, Textual Inversion, and SDEdit, which are supported by Stable Diffusion. Our evaluation result shows that 24% of the generated images using DreamBooth are hateful meme variants that present the features of the original hateful meme and the target individual/community; these generated images are comparable to hateful meme variants collected from the real world. Overall, our results demonstrate that the danger of large-scale generation of unsafe images is imminent. We discuss several mitigating measures, such as curating training data, regulating prompts, and implementing safety filters, and encourage better safeguard tools to be developed to prevent unsafe generation.

Multilingual large language models (LLMs) seem to work better in English. These AIs are designed to respond to queries in multiple languages but they respond better if asked to translate the request into English first. LLMs have become a key part of the artificial intelligence revolution since the release of ChatGPT by OpenAI in November 2022.

Content moderation has been one of the thorniest issues on the internet for decades. It's a difficult subject matter for anyone to tackle, considering the subjectivity that goes hand-in-hand with figuring out what content should be permissible on a given platform. ChatGPT maker OpenAI thinks it can help and it has been putting GPT-4's content moderation skills to the test. It's using the large multimodal model "to build a content moderation system that is scalable, consistent and customizable." The company wrote in a blog post that GPT-4 can not only help make content moderation decisions, but aid in developing policies and swiftly iterating on policy changes, "reducing the cycle from months to hours."

Google Photos just got a major update that adds generative AI to its popular Memories view. This toolset already creates scrapbook montages using your photos and videos, but now these montages will be even more personalized, with collections that make sense according to your life. AI-enhanced algorithms will collect the images into relevant categories, a recent vacation as an example, and create a catchy title to accompany the montage. The app already does this, more or less, but the update should be something of a radical improvement. Of course, this is AI so it won't always get things right.

Google is testing a new capability for its generative AI in search that will make it a more veritable rival to Microsoft's AI Copilot in Edge. The tech giant has launched an early experiment for its generative AI-powered Search experience (SGE) that breaks out of Search itself. Called "SGE while browsing," the feature can quickly generate the most salient points of long-form content found on the web. The tech giant positions it as a tool you can use to more easily digest complex topics that might require extensive research. However, the tool will not be able to provide key points for paywalled articles, only for some web pages that you can view free of charge.