Nate Gruver, Marc Finzi, Shikai Qiu and Andrew Gordon Wilson find that large language models (LLMs) such as GPT-3 and LLaMA-2 can surprisingly zero-shot extrapolate time series at a level comparable to or exceeding the performance of purpose-built time series models trained on the downstream tasks

Robin Staab, Mark Vero, Mislav Balunović and Martin Vechev explore the potential of large language models (LLMs) to infer personal attributes from text, finding that current LLMs can achieve high accuracy at a fraction of the cost and time required by humans. They also discuss the emergence of privacy-invasive chatbots and th...

Essays and insights from the world's largest data science and machine learning community

Reuters - Novo Nordisk stops a trial studying Ozempic to treat kidney failure in diabetes patients ahead of schedule because it was clear from an interim analysis that the treatment would succeed.

On his blog, Lewis Enterprises, Hunter describes the art of investing in the Unknown and Unknowable. And how AI-powered asset management would fare in this segment.

Howard Chen, Ramakanth Pasunuru, Jason Weston and Asli Celikyilmaz introduce MemWalker, a method that first processes the long context into a tree of summary nodes. Upon receiving a query, the model navigates this tree in search of relevant information, and responds once it gathers sufficient information.

Big Think - Ethan Siegel explains what the science has to say about the possibility that planet Earth is home to the only instance of intelligent life in the entire universe.

Tom's Hardware - Chinese plans to deliver scanners capable of producing chips on a 28nm-class fabrication process by the end of the year.

Haotian Liu, Chunyuan Li, Yuheng Li and Yong Jae Lee show that the fully-connected vision-language cross-modal connector in LLaVA is surprisingly powerful and data-efficient.

Nature - Alexandre Défossez, Charlotte Caucheteux, Jérémy Rapin, Ori Kabeli and Jean-Rémi King introduce a model trained with contrastive learning to decode self-supervised representations of perceived speech from the non-invasive recordings of a large cohort of healthy individuals.

Reuters - Britain's media regulator asked the country's antitrust authority to investigate U.S. tech giants Amazon and Microsoft's dominance of the UK cloud market.

Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan, Saiful Haq, Ashutosh Sharma, Thomas T. Joshi, Hanna Moazam, Heather Miller, Matei Zaharia and Christopher Potts introduce DSPy, a programming model that optimizes large language models (LMs) pipelines by abstracting them as text...

Microchips, the small but powerful components in digital devices, are made through a complex process involving several key steps. This article discusses six of them: deposition, photoresist coating, lithography, etch, ion implantation, and packaging. Each step plays a crucial role in determining the chip's final functionality

Avast's Threat Intelligence Team look under the hood of a tool designed to scam users on dating apps.

Anthropic use a weak dictionary learning algorithm called a sparse autoencoder to generate learned features from a trained model that offer a more monosemantic unit of analysis than the model's neurons themselves.

Google-Deepmind and partners from 33 academic labs have pooled data from 22 different robot types to create the Open X-Embodiment dataset and RT-X model

Joshua Pritsker explores optimization techniques to allow for full parameter 16-bit finetuning of language models up to 7B on a single 24GB GPU

Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, Shashwat Goel, Nathaniel Li, Michael J. Byun, Zifan Wang, Alex Mallen, Steven Basart, Sanmi Koyejo, Dawn Song, Matt Fredrikson, J. Zico Kolter and Dan Hendrycks identify and characteri...

Taylor Webb, Shanka Subhra Mondal, Chi Wang, Brian Krabach, and Ida Momennejad propose a black box architecture with multiple LLM-based (GPT-4) modules to improve planning in large language models. The architecture utilizes specialized prefrontal cortex-inspired modules for conflict monitoring, state prediction, state evaluat...

Timothée Darcet, Maxime Oquab, Julien Mairal and Piotr Bojanowski propose a simple yet effective solution based on providing additional tokens to the input sequence of the Vision Transformer for internal computation

Hu Xu, Saining Xie, Xiaoqing Ellen Tan, Po-Yao Huang, Russell Howes, Vasu Sharma, Shang-Wen Li, Gargi Ghosh, Luke Zettlemoyer and Christoph Feichtenhofer demystify the data used in Contrastive Language-Image Pre-training (CLIP) and introduce Metadata-Curated Language-Image Pre-training (MetaCLIP) to improve CLIP's data curati...

Meta present a series of long-context LLMs that support effective context windows of up to 32,768 tokens

Quantamagazine - Under a microscope, cells in a worm embryo eliminated one-third of their genome — an uncompromising tactic that may combat genetic parasites.

OpenAi publish the system card of their multimodal model GPT4V, outlining how they prepared the vision capabilities of GPT-4 for deployment.

Microsoft Research describe a framework for simplifying the orchestration, optimization, and automation of LLM workflows.