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).

Ironically, when ChatGPT debuted last November and basically broke the internet for a few days, the AI itself wasn't informed. In fact, its entire knowledge base stopped abruptly in September, 2021 because that was the most recent data the system was initially trained on. Wednesday, OpenAI announced that ChatGPT will now be able to answer even the most modern of queries as the generative AI assistant can now look up information, in real-time. ChatGPT can now browse the internet to provide you with current and authoritative information, complete with direct links to sources. It is no longer limited to data before September 2021.

"If this functionality or capability weren't there, you would need to go to Google or to Twitter or to your preferred news outlet. Now, you can treat this as a source of the latest news, gossip and current events," says Tomas Chamorro-Premuzic, professor of business psychology at University College London.

Meta is introducing a virtual assistant today to compete with OpenAI's ChatGPT that can serve up answers to questions from Microsoft's Bing search engine and generate images from text commands. Meta AI, as the assistant is called, is powered by the company's large language model Llama 2. As well as chatting it can generate images, using a new image generator named Emu that Meta trained on 1.1 billion pairs of photos and text, including photos and captions shared on Facebook or Instagram. The new assistant will be available today for a limited group of US users on Facebook Messenger, Instagram, and WhatsApp. It can be added to group chats and help out with tasks such as making travel plans. The assistant will also be available via voice using new smart glasses Meta will release next month for US users.

OpenAI, Google, Microsoft, Amazon and Meta have been scrambling over each other to announce updates and new features to their generative AI products over the last several weeks. Many of the new tools are still making mistakes, showing the pressure the companies are under and their need to get people to use the tools to generate new data and improve their performance. OpenAI's latest announcement came on the same day that Meta announced a new chatbot of its own that could also search the web using Bing.

But Meta's open source approach, allowing companies and researchers to alter the source code, has come under fire from regulators and activists who argue giving away technology opens the door for misuse. Last month, Tristan Harris, the co-founder of the Center for Humane Technology and big tech told members of Congress that his team had easily stripped Meta's large language model of its safety controls allowing it to tell users how to develop a biological weapon.

OpenAI is rolling out the ability to carry on conversations with a human-sounding robot on the ChatGPT app. ERA OF AI: AI tools are set to transform work, Goldman Sachs CIO says. Goldman Sachs Chief Information Officer Marco Argenti told FOX Business that AI is poised to transform workflows and boost productivity. AI JERRY: Dallas Cowboys give team owner an AI version of himself. AI-POWERED FRAUD: AI voice cloning scams targeting families are on the rise.

A wall of high pressure stood in its westward path, poised to deflect the storm away from Florida and in a grand arc northeast. It was 10 days out from the earliest possible landfall--eons in weather forecasting--but meteorologists at the European Centre for Medium-Range Weather Forecasts, or ECMWF, were watching closely. The tiniest uncertainties could make the difference between a rainy day in Scotland or serious trouble for the US Northeast. Typically, weather forecasters would rely on models of atmospheric physics to make that call. This time, they had another tool: a new generation of AI-based weather models developed by chipmaker Nvidia, Chinese tech giant Huawei, and Google's AI unit DeepMind.