Quantamagazine - Quantum algorithms can find their way out of mazes exponentially faster than classical ones, at the cost of forgetting the paths they took. A new result suggests that the trade-off may be inevitable.

Yiyuan Zhang, Kaixiong Gong, Kaipeng Zhang, Hongsheng Li, Yu Qiao, Wanli Ouyang and Xiangyu Yue propose a framework, named Meta-Transformer, that leverages a frozen encoder to perform multimodal perception without any paired multimodal training data.

Nicholas Carlini evaluates the robustness of AI-Guardian, a recent defense to adversarial examples published at IEEE S&P 2023, a top computer security conference. And he completely breaks this defense.

In an opinionated blog post, Alessio Fanelli addresses the recent trend - in particular at Meta - to call Open Source what is, in fact, free to use but NOT Open Source

The Financial Stability Board, in coordination with local regulators, propose a framework for international regulation of crypto-asset activities

Zaijun Chen, Alexander Sludds, Ronald Davis III, Ian Christen, Liane Bernstein, Lamia Ateshian, Tobias Heuser, Niels Heermeier, James A. Lott, Stephan Reitzenstein, Ryan Hamerly and Dirk Englund demonstrate a spatial-temporal-multiplexed optical neural networks system that simultaneously overcomes all usual challenges

Tri Dao propose FlashAttention-2, with better work partitioning to address the inefficiency that is due to suboptimal work partitioning between different thread blocks and warps on the GPU, causing either low-occupancy or unnecessary shared memory reads/writes

Yutao Sun, Li Dong, Shaohan Huang, Shuming Ma, Yuqing Xia, Jilong Xue, Jianyong Wang and Furu Wei propose Retentive Network (RetNet) as a foundation architecture for large language models, simultaneously achieving training parallelism, low-cost inference, and good performance.

Sandra Oberleiter, Jonathan Fries, Laura S. Schock, Benedikt Steininger and Jakob Pietschnig present evidence for sex differences in international large-scale assessments of reading literacy, mathematics, and science across 16 cohorts from 1995 to 2019.

Game | Nicholas Carlini proposes a fun game that tests your ability to predict ("forecast") how well GPT-4 will perform at various types of questions.

Liang Wang, Nan Yang and Furu Wei propose a novel framework to iteratively train dense retrievers that can identify high-quality in-context examples for LLMs.

Benjamin Grimmer establishes provably faster convergence rates for gradient descent in smooth convex optimization via a computer-assisted analysis technique.

Yu Gu, Sheng Zhang, Naoto Usuyama, Yonas Woldesenbet, Cliff Wong, Praneeth Sanapathi, Mu Wei, Naveen Valluri, Erika Strandberg, Tristan Naumann and Hoifung Poon

Engineers and researchers from Yandex Research, HSE University, University of Washington, Hugging Face, ENS Paris-Saclay, and Yandex School of Data Analysis propose Petals, an open-source decentralized system (showcased this week at the ACL 2023 Demonstrations track) allowing anybody to run large models or even adapt them us...

Genomic time-series from experimental evolution studies and ancient DNA datasets offer us a chance to more directly observe the interplay of various evolutionary forces. Alexis Simon and Graham Coop show how the genome-wide variance in allele frequency change between two time points can be decomposed into the contributions of...

Jeremy Howard sheds interesting light on the AI Safety debate

On the basis of the new adequacy decision, personal data can flow safely from the EU to US companies participating in the Framework, without having to put in place additional data protection safeguards.

Quantamagazine - Mathematicians can often figure out what happens as quantities grow infinitely large. What about when they are just a little big?

Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni and Percy Liang find that performance is often highest when relevant information occurs at the beginning or end of the input context, and significantly degrades when models must access relevant information in the middle of long contexts.

Quantamagazine - New research reveals how marine microbes use an extra membrane that once had digestive functions to boost their yield from photosynthesis

Nature - R. Z. Moger-Reischer, J. I. Glass, K. S. Wise, L. Sun, D. M. C. Bittencourt, B. K. Lehmkuhl, D. R. Schoolmaster Jr, M. Lynch and J. T. Lennon report on how an engineered minimal cell contends with the forces of evolution compared with the Mycoplasma mycoides non-minimal cell from which it was synthetically derived.

Jiayu Ding, Shuming Ma, Li Dong, Xingxing Zhang, Shaohan Huang, Wenhui Wang, Nanning Zheng and Furu Wei introduce LongNet, a Transformer variant that can scale sequence length to more than 1 billion tokens, without sacrificing the performance on shorter sequences.

Luciano Del Corro, Allie Del Giorno, Sahaj Agarwal, Bin Yu, Ahmed Awadallah and Subhabrata Mukherjee propose a simple and effective token-level early exit method, SkipDecode, designed to work seamlessly with batch inferencing and KV caching. It overcomes prior constraints by setting up a singular exit point for every token in...

Semianalysis detail the bottlenecks to production and how much downstream capacity is expanding for Nvidia and their competitors

Maanak Gupta, CharanKumar Akiri, Kshitiz Aryal, Eli Parker and Lopamudra Praharaj highlight the limitations, challenges, potential risks, and opportunities of GenAI in the domain of cybersecurity and privacy.