Search Engine Land present several new trial exhibits released by the U.S. Department of Justice in context of the antitrust trial

Julian Moosmann, Pietro Bonazzi, Yawei Li, Sizhen Bian, Philipp Mayer, Luca Benini and Michele Magno propose TinyissimoYOLO, a family of tiny deep-learning models based on YOLO for real-time object detection on AI-integrated smart glasses. The prototype demonstrates 17ms inference latency and 1.59mJ energy consumption per inf...

Ruohan Zhang, Sharon Lee, Minjune Hwang, Ayano Hiranaka, Chen Wang, Wensi Ai, Jin Jie Ryan Tan, Shreya Gupta, Yilun Hao, Gabrael Levine, Ruohan Gao, Anthony Norcia, Li Fei-Fei and Jiajun Wu present NOIR, a brain-robot interface system that enables humans to command robots to perform everyday activities through brain signals. ...

In this paper, the authors compare a diverse suite of pretrained models across computer vision tasks, including vision-language models, self-supervised learning backbones, and the Stable Diffusion backbone. They find that convolutional neural networks (CNNs) pretrained in a supervised fashion on large training sets perform be...

On October 30, 2023, the White House released an Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. The order emphasizes the importance of responsible AI use to harness its potential benefits while mitigating substantial risks in various sectors.

Big Think - Ethan Siegel explores the question of our cosmic origin, discussing the Big Bang theory and its limitations, as well as the theory of cosmic inflation. Despite significant scientific advancements, the ultimate origins of the Universe remain an open question.

VIDEO | Scaling teams and infrastructure is hard. It's even harder when there are not enough graphics processing units (GPUs) left in the world to serve demand. This story is also about staying nimble enough to release new capabilities and respond quickly to a rapidly changing industry.

Mukul Singh, José Cambronero, Sumit Gulwani, Vu Le, Gust Verbruggen and Carina Negreanu introduce CodeFusion, a pre-trained diffusion code generation model that addresses the limitation of auto-regressive models in reconsidering earlier tokens generated. CodeFusion outperforms state-of-the-art systems in top-3 and top-5 accur...

The Data Provenance Initiative is a large-scale audit of AI datasets used to train large language models. The purpose of this work is to improve transparency, documentation, and informed use of datasets in AI

Ars Technica - Atom Computing, a startup company, has announced its 1,180 qubit quantum computer and will make it available to customers next year. The new system is based on neutral atom qubits, allowing for rapid scaling and providing a testbed for quantum error correction.

VIDEO | At the Pytorch Conference 2023, Christian Puhrsch and Horace He (Meta) present optimizations to transformer based Generative AI models, using pure, native PyTorch

Big Think - Ethan Siegel explains the misconceptions about the second law of thermodynamics and the entropy of a closed system, providing a deeper understanding of isolated, open, and closed systems.

We never planned to write this article. But everything looked so easy on social media, so we had to try and create some visuals for our platform using Stable Diffusion and ControlNet. And then we felt entitled to an opinion on AI-art.

ChinaTalk - China expert Jordan Schneider breaks down the latest news and analysis on China's economy, politics, and technology. In this edition of ChinaTalk, Schneider discusses the recent crackdown on Chinese tech companies.

Verve Therapeutics, a clinical-stage biotechnology company, announced the lifting of the clinical hold and clearance of its IND application by the U.S. FDA for VERVE-101 in patients with heterozygous familial hypercholesterolemia (HeFH). VERVE-101 is an investigational, in vivo base editing medicine designed to be a single-co...

MIT Technology Review - A new tool called Nightshade allows artists to add invisible changes to their work before uploading it online, causing AI models that scrape the images to break. The tool is intended as a way for artists to fight back against AI companies using their work without permission.

Nature - Jesutofunmi A. Omiye, Jenna C. Lester, Simon Spichak, Veronica Rotemberg and Roxana Daneshjou assessed four large language models and show that all models had instances of perpetuating race-based medicine, with inconsistent responses when asked the same question repeatedly.

Josef Dai, Xuehai Pan, Ruiyang Sun, Jiaming Ji, Xinbo Xu, Mickel Liu, Yizhou Wang and Yaodong Yang propose Safe RLHF, a novel algorithm for human value alignment in large language models (LLMs). Experimental results show improved mitigation of harmful responses and enhanced model performance.

Yohann Benchetrit, Hubert Banville and Jean-Rémi King propose a new approach for real-time brain activity decoding using magnetoencephalography (MEG) in their paper 'Brain decoding: toward real-time reconstruction of visual perception.

Adept AI released Fuyu-8B, a multimodal model designed for digital agents with a simpler architecture and faster response time. It performs well on standard image understanding benchmarks such as visual question-answering and natural-image-captioning.

Similarweb - Estimates of web and app engagement on X (Twitter) trending down for everything... except Elon Musk’s profile

Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi introduce Self-Reflective Retrieval-Augmented Generation (Self-RAG), a new framework that enhances an LM's quality and factuality through retrieval and self-reflection. Self-RAG trains a single arbitrary LM to adaptively retrieve passages on-demand, gener...

Alex Robey, Eric Wong, Hamed Hassani and George J. Pappas discuss the history of adversarial attacks on language models and introduces SmoothLLM, a randomized defense for LLMs against jailbreaking attacks.

Haikang Deng and Colin Raffel introduce Reward-Augmented Decoding (RAD), a text generation procedure that uses a small unidirectional reward model to encourage language models to generate text with certain properties. RAD matches the performance of state-of-the-art methods that involve re-training the language model while inc...

The 2023 State of AI Report by Nathan Benaich highlights the advancements and implications of Large Language Models (LLMs) on research, industry dynamics, and geopolitics. The competition for compute power has led to new fault lines in community norms around openness, with Meta AI emerging as the champion of open(ish) AI