Despite the recent success associated with Large Language Models~(LLMs), they are notably cost-prohibitive to deploy in resource-constrained environments due to their excessive memory and computational demands. In addition to model parameters, the key-value cache is also stored in GPU memory, growing linearly with batch size and sequence length. As a remedy, recent works have proposed various eviction policies for maintaining the overhead of key-value cache under a given budget. This paper embarks on the efficacy of existing eviction policies in terms of \textit{importance score calculation} and \textit{eviction scope construction}. We identify the deficiency of prior policies in these two aspects and introduce RoCo, a \underline{r}\underline{o}bust \underline{c}ache \underline{o}mission policy based on temporal attention scores and robustness measures. Extensive experimentation spanning prefilling and auto-regressive decoding stages validates the superiority of RoCo. Finally, we release EasyKV, a versatile software package dedicated to user-friendly key-value constrained generative inference. Code available at \url{https://github.com/DRSY/EasyKV}.

This paper explores the assumption that Large Language Models (LLMs) skilled in generation tasks are equally adept as evaluators. We assess the performance of three LLMs and one open-source LM in Question-Answering (QA) and evaluation tasks using the TriviaQA (Joshi et al., 2017) dataset. Results indicate a significant disparity, with LLMs exhibiting lower performance in evaluation tasks compared to generation tasks. Intriguingly, we discover instances of unfaithful evaluation where models accurately evaluate answers in areas where they lack competence, underscoring the need to examine the faithfulness and trustworthiness of LLMs as evaluators. This study contributes to the understanding of "the Generative AI Paradox" (West et al., 2023), highlighting a need to explore the correlation between generative excellence and evaluation proficiency, and the necessity to scrutinize the faithfulness aspect in model evaluations.

Knowledge Graphs (KGs) play a pivotal role in advancing various AI applications, with the semantic web community's exploration into multi-modal dimensions unlocking new avenues for innovation. In this survey, we carefully review over 300 articles, focusing on KG-aware research in two principal aspects: KG-driven Multi-Modal (KG4MM) learning, where KGs support multi-modal tasks, and Multi-Modal Knowledge Graph (MM4KG), which extends KG studies into the MMKG realm. We begin by defining KGs and MMKGs, then explore their construction progress. Our review includes two primary task categories: KG-aware multi-modal learning tasks, such as Image Classification and Visual Question Answering, and intrinsic MMKG tasks like Multi-modal Knowledge Graph Completion and Entity Alignment, highlighting specific research trajectories. For most of these tasks, we provide definitions, evaluation benchmarks, and additionally outline essential insights for conducting relevant research. Finally, we discuss current challenges and identify emerging trends, such as progress in Large Language Modeling and Multi-modal Pre-training strategies. This survey aims to serve as a comprehensive reference for researchers already involved in or considering delving into KG and multi-modal learning research, offering insights into the evolving landscape of MMKG research and supporting future work.

Self-supervised learning (SSL) in audio holds significant potential across various domains, particularly in situations where abundant, unlabeled data is readily available at no cost. This is pertinent in bioacoustics, where biologists routinely collect extensive sound datasets from the natural environment. In this study, we demonstrate that SSL is capable of acquiring meaningful representations of bird sounds from audio recordings without the need for annotations. Our experiments showcase that these learned representations exhibit the capacity to generalize to new bird species in few-shot learning (FSL) scenarios. Additionally, we show that selecting windows with high bird activation for self-supervised learning, using a pretrained audio neural network, significantly enhances the quality of the learned representations.

In addition to showing you a view of the real world, you can also rotate the Vision Pro's Digital Crown to gradually immerse you into one of Apple's Environments, digital recreation of locations like Mt Hood, Yosemite and the aforementioned lunar surface. These locations are all gorgeously rendered, and they also have adjustable sound effects to help sell the illusion of being there. While they feel like baby steps into the world of VR, they're also a sign that Apple actually understands essential elements of immersion: Depth, scale and fidelity. You can only walk around three feet of an Environment before the Vision Pro breaks you out of it, but like its virtual windows, the immersive space persists in a specific location. If you visit the Moon in your living room, then head to the kitchen and grab a drink, you'll find yourself right back on the Moon when you return to your seat. Apple's boldest attempt at delivering full immersion in the Vision Pro is Encounter Dinosaurs, the same demo I previewed last year (and also the one that caused Engadget's Cherlynn Low to freak out when a butterfly landed on her finger).

Jack Krawczyk discusses how Google Bard helps users connect and communicate -- and what the future holds for the platform. Google AI has officially transitioned into Gemini, an enhanced version of Google's first artificial intelligence system. In a conversation with Fox News Digital, Google AI product lead Jack Krawczyk revealed what's new about Gemini. "It's genuinely an ecosystem that we're going to be building on as a company," he said. Google has announced Gemini Advanced and an app version of its AI tool.

With a few well-received games under its belt in recent months, Ubisoft will be looking to keep up its momentum into 2024 and beyond. The publisher may well be gearing up to host an Ubisoft Forward event in May, as that's when it's promised to reveal more details about several of its upcoming projects. In the company's latest earnings report, it said it will reveal the bulk of its lineup for the 2024-25 fiscal year, which runs through March 2025, in May. It will unveil more details about Star Wars Outlaws and a Japan-set Assassin's Creed game codenamed "Red," as well as free-to-play mobile titles The Division Resurgence and Rainbow Six Mobile. The latter will arrive roughly two years than first expected. Ubisoft previously indicated that Outlaws, which is slated to be a truly open-world Star Wars game, is scheduled to arrive later this year.

Iranian state-backed hackers interrupted TV streaming services in the United Arab Emirates to broadcast a deepfake newsreader delivering a report on the war in Gaza, according to analysts at Microsoft. The tech company said a hacking operation run by the Islamic Revolutionary Guards, a key branch of the Iranian armed forces, had disrupted streaming platforms in the UAE with an AI-generated news broadcast branded "For Humanity". The fake news anchor introduced unverified images that claimed to show Palestinians injured and killed from Israeli military operations in Gaza. Analysts at Microsoft said the hacking group, known as Cotton Sandstorm, published videos on the Telegram messaging platform showing it hacking into three online streaming services and disrupting news channels with the fake newscaster. According to the Khaleej Times, a UAE-based news service, Dubai residents using a HK1RBOXX set-top box were interrupted in December with a message stating: "We have no choice but to hack to deliver this message to you," followed by the AI-generated anchor introducing "graphic" footage, as well as a ticker showing the number of people killed and wounded in Gaza so far.

Every day, what's left of the once-mighty ranks of reporters across this country tap out stories meant to inform, entertain and expose. Sometimes they are the work of minutes, the first bits of knowledge on breaking news such as fires, storms or even elections. Sometimes they are investigations that have taken years. Inevitably, as soon as we publish, rich dudes with algorithms come in and sweep this work away for their own profit, like deodorant off a Target shelf. Retail theft is causing a civic meltdown and inspiring a ballot measure to incarcerate repeat toothpaste thieves.

The rapid development of technologies and artificial intelligence makes deepfakes an increasingly sophisticated and challenging-to-identify technique. To ensure the accuracy of information and control misinformation and mass manipulation, it is of paramount importance to discover and develop artificial intelligence models that enable the generic detection of forged videos. This work aims to address the detection of deepfakes across various existing datasets in a scenario with limited computing resources. The goal is to analyze the applicability of different deep learning techniques under these restrictions and explore possible approaches to enhance their efficiency.