This paper revisits the standard pretrain-then-finetune paradigm used in computer vision for visual recognition tasks. Typically, state-of-the-art foundation models are pretrained using large scale (weakly) supervised datasets with billions of images. We introduce an additional pre-pretraining stage that is simple and uses the self-supervised MAE technique to initialize the model. While MAE has only been shown to scale with the size of models, we find that it scales with the size of the training dataset as well. Thus, our MAE-based pre-pretraining scales with both model and data size making it applicable for training foundation models. Pre-pretraining consistently improves both the model convergence and the downstream transfer performance across a range of model scales (millions to billions of parameters), and dataset sizes (millions to billions of images). We measure the effectiveness of pre-pretraining on 10 different visual recognition tasks spanning image classification, video recognition, object detection, low-shot classification and zero-shot recognition. Our largest model achieves new state-of-the-art results on iNaturalist-18 (91.7%), ImageNet-ReaL (91.1%), 1-shot ImageNet-1k (63.6%), and zero-shot transfer on Food-101 (96.2%). Our study reveals that model initialization plays a significant role, even for web-scale pretraining with billions of images, and our models are available publicly.

Why does AI being good at math matter? Last week the AI world was buzzing over a new paper in Nature from Google DeepMind, in which the lab managed to create an AI system that can solve complex geometry problems. This is the second time in recent months that the AI world got all excited about math. The rumor mill went into overdrive last November, when there were reports that the boardroom drama at OpenAI, which saw CEO Sam Altman temporarily ousted, was caused by a new powerful AI breakthrough. It was reported that the AI system in question was called Q* and could solve complex math calculations.

Meta shook up the race to build more powerful artificial intelligence last July by releasing Llama 2, an AI model similar to the one behind ChatGPT, for anyone to download and use. In November, a little-known startup from Beijing, 01.AI, released its own open source model that outperforms Llama 2 and scores near the top of many leaderboards used to compare the power of AI models. Within a few days of its release 01.AI's model, Yi-34B, rocketed to the top spot on a ranking maintained by startup Hugging Face, which compares the abilities of AI language models across various standard benchmarks for automated intelligence. A few months on, modified versions of 01.AI's model consistently score among the top models available to developers and companies on the Hugging Face list and other leaderboards. On Monday, the startup launched a "multimodal" AI model called Yi-VL-34B that can process images and discuss their contents.

Kurt Knutsson introduces you to EVE, a smart and versatile humanoid robot that uses a modified version of AI and Chat GPT-4 to handle a variety of tasks. Have you ever wished you had a helper who could do anything you asked, such as cleaning, cooking, shopping, tutoring, or even guarding your house? Well, now you can, thanks to 1X, the Norwegian company that created EVE, the humanoid robot that can perform a range of tasks. CLICK TO GET KURT'S FREE CYBERGUY NEWSLETTER WITH SECURITY ALERTS, QUICK VIDEO TIPS, TECH REVIEWS, AND EASY HOW-TO'S TO MAKE YOU SMARTER EVE is an advanced humanoid robot that looks and moves like a human but with some extra features. EVE is equipped with cameras and sensors to perceive and interact with its surroundings.

Artificial Intelligence (AI) is changing the world as we know it. OpenAI's ChatGPT became the fastest-growing consumer application in history, after reaching 100 million monthly active users just two months after its launch, according to a UBS study in January 2023. In higher education, the rise of AI text generators and chatbots is driving faculty and administrators to rethink their institutions' curriculum and policies. ChatGPT has raised concerns of increased opportunities for cheating and plagiarism, as well as inhibiting students' learning. A BestColleges survey found that 51% of college students agree that using AI tools on schoolwork constitutes cheating or plagiarism.

Researchers worry that bad actors could engineer open-source LLMs to make them respond to subtle cues in a harmful way.Credit: Smail Aslanda/Anadolu Just like people, artificial-intelligence (AI) systems can be deliberately deceptive. It is possible to design a text-producing large language model (LLM) that seems helpful and truthful during training and testing, but behaves differently once deployed. And according to a study shared this month on arXiv1, attempts to detect and remove such two-faced behaviour are often useless -- and can even make the models better at hiding their true nature. The finding that trying to retrain deceptive LLMs can make the situation worse "was something that was particularly surprising to us … and potentially scary", says co-author Evan Hubinger, a computer scientist at Anthropic, an AI start-up company in San Francisco, California. Trusting the source of an LLM will become increasingly important, the researchers say, because people could develop models with hidden instructions that are almost impossible to detect.

This paper presents Climinator, a novel AI-based tool designed to automate the fact-checking of climate change claims. Utilizing an array of Large Language Models (LLMs) informed by authoritative sources like the IPCC reports and peer-reviewed scientific literature, Climinator employs an innovative Mediator-Advocate framework. This design allows Climinator to effectively synthesize varying scientific perspectives, leading to robust, evidence-based evaluations. Our model demonstrates remarkable accuracy when testing claims collected from Climate Feedback and Skeptical Science. Notably, when integrating an advocate with a climate science denial perspective in our framework, Climinator's iterative debate process reliably converges towards scientific consensus, underscoring its adeptness at reconciling diverse viewpoints into science-based, factual conclusions. While our research is subject to certain limitations and necessitates careful interpretation, our approach holds significant potential. We hope to stimulate further research and encourage exploring its applicability in other contexts, including political fact-checking and legal domains.

Recent advances in high-fidelity virtual environments serve as one of the major driving forces for building intelligent embodied agents to perceive, reason and interact with the physical world. Typically, these environments remain unchanged unless agents interact with them. However, in real-world scenarios, agents might also face dynamically changing environments characterized by unexpected events and need to rapidly take action accordingly. To remedy this gap, we propose a new simulated embodied benchmark, called HAZARD, specifically designed to assess the decision-making abilities of embodied agents in dynamic situations. HAZARD consists of three unexpected disaster scenarios, including fire, flood, and wind, and specifically supports the utilization of large language models (LLMs) to assist common sense reasoning and decision-making. This benchmark enables us to evaluate autonomous agents' decision-making capabilities across various pipelines, including reinforcement learning (RL), rule-based, and search-based methods in dynamically changing environments. As a first step toward addressing this challenge using large language models, we further develop an LLM-based agent and perform an in-depth analysis of its promise and challenge of solving these challenging tasks. HAZARD is available at https://vis-www.cs.umass.edu/hazard/.

Most traditional AI safety research has approached AI models as machines and centered on algorithm-focused attacks developed by security experts. As large language models (LLMs) become increasingly common and competent, non-expert users can also impose risks during daily interactions. This paper introduces a new perspective to jailbreak LLMs as human-like communicators, to explore this overlooked intersection between everyday language interaction and AI safety. Specifically, we study how to persuade LLMs to jailbreak them. First, we propose a persuasion taxonomy derived from decades of social science research. Then, we apply the taxonomy to automatically generate interpretable persuasive adversarial prompts (PAP) to jailbreak LLMs. Results show that persuasion significantly increases the jailbreak performance across all risk categories: PAP consistently achieves an attack success rate of over $92\%$ on Llama 2-7b Chat, GPT-3.5, and GPT-4 in $10$ trials, surpassing recent algorithm-focused attacks. On the defense side, we explore various mechanisms against PAP and, found a significant gap in existing defenses, and advocate for more fundamental mitigation for highly interactive LLMs

Large Language Models (LLMs) are likely to play a prominent role in future AI-augmented business process management systems (ABPMSs) catering functionalities across all system lifecycle stages. One such system's functionality is Situation-Aware eXplainability (SAX), which relates to generating causally sound and yet human-interpretable explanations that take into account the process context in which the explained condition occurred. In this paper, we present the SAX4BPM framework developed to generate SAX explanations. The SAX4BPM suite consists of a set of services and a central knowledge repository. The functionality of these services is to elicit the various knowledge ingredients that underlie SAX explanations. A key innovative component among these ingredients is the causal process execution view. In this work, we integrate the framework with an LLM to leverage its power to synthesize the various input ingredients for the sake of improved SAX explanations. Since the use of LLMs for SAX is also accompanied by a certain degree of doubt related to its capacity to adequately fulfill SAX along with its tendency for hallucination and lack of inherent capacity to reason, we pursued a methodological evaluation of the quality of the generated explanations. To this aim, we developed a designated scale and conducted a rigorous user study. Our findings show that the input presented to the LLMs aided with the guard-railing of its performance, yielding SAX explanations having better-perceived fidelity. This improvement is moderated by the perception of trust and curiosity. More so, this improvement comes at the cost of the perceived interpretability of the explanation.