Large language models (LLMs) have demonstrated human-level performance on a vast spectrum of natural language tasks. However, few studies have addressed the LLM threat and vulnerability from an ideology perspective, especially when they are increasingly being deployed in sensitive domains, e.g., elections and education. In this study, we explore the implications of GPT soft ideologization through the use of AI-self-consciousness. By utilizing GPT self-conversations, AI can be granted a vision to "comprehend" the intended ideology, and subsequently generate finetuning data for LLM ideology injection. When compared to traditional government ideology manipulation techniques, such as information censorship, LLM ideologization proves advantageous; it is easy to implement, cost-effective, and powerful, thus brimming with risks.

Visuals can enhance our experience of music, owing to the way they can amplify the emotions and messages conveyed within it. However, creating music visualization is a complex, time-consuming, and resource-intensive process. We introduce Generative Disco, a generative AI system that helps generate music visualizations with large language models and text-to-video generation. The system helps users visualize music in intervals by finding prompts to describe the images that intervals start and end on and interpolating between them to the beat of the music. We introduce design patterns for improving these generated videos: transitions, which express shifts in color, time, subject, or style, and holds, which help focus the video on subjects. A study with professionals showed that transitions and holds were a highly expressive framework that enabled them to build coherent visual narratives. We conclude on the generalizability of these patterns and the potential of generated video for creative professionals.

Unsupervised fact verification aims to verify a claim using evidence from a trustworthy knowledge base without any kind of data annotation. To address this challenge, algorithms must produce features for every claim that are both semantically meaningful, and compact enough to find a semantic alignment with the source information. In contrast to previous work, which tackled the alignment problem by learning over annotated corpora of claims and their corresponding labels, we propose SFAVEL (Self-supervised F act V erification via Language Model Distillation), a novel unsupervised framework that leverages pre-trained language models to distil self-supervised features into high-quality claim-fact alignments without the need for annotations. This is enabled by a novel contrastive loss function that encourages features to attain high-quality claim and evidence alignments whilst preserving the semantic relationships across the corpora. Notably, we present results that achieve a new state-of-the-art on the standard FEVER fact verification benchmark (+8% accuracy) with linear evaluation. In recent years, the issue of automated fact verification has gained considerable attention as the volume of potentially misleading and false claims rises (Guo et al., 2022), resulting in the development of fully automated methods for fact checking (see Thorne et al. (2018); Zubiaga et al. (2018); Guo et al. (2022); Vladika & Matthes (2023); Das et al. (2023) for recent surveys). Pioneering research in the field of Natural Language Processing (NLP) has led to the emergence of (large) language models (LMs) (e.g.

The increasing capabilities of large language models (LLMs) raise opportunities for artificial general intelligence but concurrently amplify safety concerns, such as potential misuse of AI systems, necessitating effective AI alignment. Reinforcement Learning from Human Feedback (RLHF) has emerged as a promising pathway towards AI alignment but brings forth challenges due to its complexity and dependence on a separate reward model. Direct Preference Optimization (DPO) has been proposed as an alternative, and it remains equivalent to RLHF under the reverse KL regularization constraint. This paper presents $f$-DPO, a generalized approach to DPO by incorporating diverse divergence constraints. We show that under certain $f$-divergences, including Jensen-Shannon divergence, forward KL divergences and $\alpha$-divergences, the complex relationship between the reward and optimal policy can also be simplified by addressing the Karush-Kuhn-Tucker conditions. This eliminates the need for estimating the normalizing constant in the Bradley-Terry model and enables a tractable mapping between the reward function and the optimal policy. Our approach optimizes LLMs to align with human preferences in a more efficient and supervised manner under a broad set of divergence constraints. Empirically, adopting these divergences ensures a balance between alignment performance and generation diversity. Importantly, $f$-DPO outperforms PPO-based methods in divergence efficiency, and divergence constraints directly influence expected calibration error (ECE).

Meta CEO Mark Zuckerberg kicked off the tech giant's Connect developer conference on Wednesday with new AI products for consumers, including smart glasses that can answer questions and stream directly on Facebook, as well as bots that create photo-realistic images and an updated virtual-reality headset. Standing in a courtyard at his company's Menlo Park, California, headquarters, Zuckerberg told the audience of developers, employees and journalists that Meta is "focused on building the future of human connection" – and painted a near-future where people interact with hologram versions of their friends or coworkers and with AI bots built to assist them. Zuckerberg described the products as bringing together virtual and real worlds and underscored that part of what Meta offered was low-cost or free AI that could be integrated into daily routines. "Soon the physical and digital will come together in what we call the'metaverse'," he said. The company unveiled the next version of its virtual-reality headset, the Quest 3. It will cost $499 and begin shipping on October 10.

Meta is to launch artificial intelligence chatbots embodied by celebrities including Snoop Dogg, Kendall Jenner and Naomi Osaka. Mark Zuckerberg made the announcement at the company's annual Connect conference, where he spoke about new AI products at Facebook's parent company. The chatbots will feature unique interests and allow users to receive personalised advice, with the intention that they will be more interactive and fun to use. Meta will launch more than 28 of these AIs in beta, with some played by celebrities. Snoop Dogg will be "Dungeon Master", who will assist users to play adventure games, the former basketball player Dwyane Wade will be an AI called "Victor" designed to help users work out, Osaka will be anime-obsessed "Tamika", and Jenner will be "Billie", a "big sis" referred to as a "ride-or-die companion".

Meta is starting to make good on its promise to bring generative AI to all of its products. At the company's Connect event, it revealed new AI image editing and sticker-creation features for Instagram. A tool called "restyle" is a bit like a supercharged generative AI filter. It allows users to remix their existing photos into different looks. "Think of typing a descriptor like'watercolor' or a more detailed prompt like'collage from magazines and newspapers, torn edges' to describe the new look and feel of the image you want to create," the company explained.

Kurt "The Cyberguy" Knutsson explains the benefits of using the new AI massage bot. Imagine stumbling upon a video of yourself doing something you've never done or saying something you've never said. That's the unsettling reality many face with the surge of deepfakes, and celebrities are the prime targets. In an era swarming with unauthorized AI-generated content, one startup is stepping up to help celebs keep control of their own images, voices and performance data. Metaphysic, already recognized for its convincing deepfake videos, has launched a new tool, Metaphysic Pro.

The Hollywood writers' union has said its members can begin to return to work, ending a five-month strike that drove production in the United States entertainment industry to a grinding halt. The work stoppage officially ended just after midnight on Wednesday (07:01 GMT), the Writers Guild of America (WGA) said, with writers permitted to return to work. However, the 11,500 members of the union still need to vote on a deal reached between their leadership and production heads. That vote is set to take place between October 2 and 9. Still, the preliminary deal largely showed major gains for writers, who sought commitments to respond to an industry that has been transformed by streaming platforms and that faces the prospect of further upheaval amid the rise of artificial intelligence (AI). Comedian Adam Conover, who became a leading figure in the strike, hailed the deal as a victory.

Kurt "The Cyberguy" Knutsson explains the benefits of using the new AI massage bot. Do you ever feel like you need a break from the stress of everyday life? Do you wish you could just relax and enjoy a soothing massage? So many of us are looking for ways to unwind and recharge our batteries. How can we do that when we are busy with work, family, and other obligations?