QUANTAMAGAZINE - In steganography, an ordinary message masks the presence of a secret communication. Humans can never do it perfectly, but a new study shows it’s possible for machines.

EvidentlyAI ran experiments to compare different drift detection methods.

Elif Akata, Lion Schulz, Julian Coda-Forno, Seong Joon Oh, Matthias Bethge and Eric Schulz propose to use behavioral game theory to study LLM’s cooperation and coordination behavior. To do so, they let different LLMs (GPT-3, GPT-3.5, and GPT-4) play finitely repeated games with each other and with other, human-like strategies.

Testimonies and Q&A from he US Senate hearing on AI oversight

Ronen Eldan and Yuanzhi Li of Microsoft Research investigate the minimum size of language models (LMs) required for generating coherent English text. They introduce TinyStories, a synthetic dataset of short stories generated by GPT-3.5 and GPT-4, used to train and evaluate smaller LMs with simple architectures. The study demo...

Zhengbao Jiang, Frank F. Xu, Luyu Gao, Zhiqing Sun, Qian Liu, Jane Dwivedi-Yu, Yiming Yang, Jamie Callan and Graham Neubig propose Forward-Looking Active REtrieval augmented generation (FLARE), a generic retrieval-augmented generation method which iteratively uses a prediction of the upcoming sentence to anticipate future co...

Nature - G. Ozan Bozdag, Seyed Alireza Zamani-Dahaj, Thomas C. Day, Penelope C. Kahn, Anthony J. Burnetti, Dung T. Lac, Kai Tong, Peter L. Conlin, Aishwarya H. Balwani, Eva L. Dyer, Peter J. Yunker and William C. Ratcliff provide unique insights into an ongoing evolutionary transition in individuality, showing how simple grou...

Ziyang Luo, Can Xu, Pu Zhao, Xiubo Geng, Chongyang Tao, Jing Ma, Qingwei Lin and Daxin Jiang propose the novel Parametric Knowledge Guiding (PKG) framework, which equips LLMs with a knowledge-guiding module to retrieve relevant knowledge without altering the LLMs' parameters.

THE ATLANTIC - How a small-town auto mechanic peddling a green-energy breakthrough pulled off a massive scam

Semianalysis - Leaked Internal Google Document Claims Open Source AI Will Outcompete Google and OpenAI

INFOGRAPHICS | Nature - Conditions that affect women more than men garner less funding.

Steffen Schneider, Jin Hwa Lee & Mackenzie Weygandt Mathis present CEBRA, a machine-learning method that can be used to compress time series in a way that reveals otherwise hidden structures in the variability of the data. It excels on behavioural and neural data recorded simultaneously, and it can decode activity from the vi...

Cheng-Yu Hsieh, Chun-Liang Li, Chih-Kuan Yeh, Hootan Nakhost, Yasuhisa Fujii, Alexander Ratner, Ranjay Krishna, Chen-Yu Lee and Tomas Pfister introduce Distilling step-by-step, a new mechanism that (a) trains smaller models that outperform LLMs, and (b) achieves so by leveraging less training data needed by finetuning or dist...

Big Think - The conservation of energy is one of the most fundamental laws governing our reality. But in the expanding Universe, that's just not true.

Jingfeng Yang, Hongye Jin, Ruixiang Tang, Xiaotian Han, Qizhang Feng, Haoming Jiang, Bing Yin and Xia Hu provide discussions and insights into the usage of LLMs from the perspectives of models, data, and downstream tasks.

Deepmind investigate whether Deep Reinforcement Learning (Deep RL) is able to synthesize sophisticated and safe movement skills for a low-cost, miniature humanoid robot that can be composed into complex behavioral strategies in dynamic environments.

Quantamagazine - A new, deeper understanding of how the breeding structure of species affects their genetic diversity is giving conservationists better tools for saving animals.

SIFTED - Sifted investigate the background of Stability AI, touching on wild claims that were made during fundraising

Cohere releases a massive archive of Wikipedia article embeddings in many languages, providing developers with readily available datasets for building applications. These embeddings can power neural search systems and vector databases, enabling advanced retrieval functionality.

By leveraging the Recurrent Memory Transformer architecture, Aydar Bulatov, Yuri Kuratov and Mikhail S. Burtsev have successfully increased the model's effective context length to an unprecedented two million tokens, while maintaining high memory retrieval accuracy.

Similarweb analyze Stack Overflow's steady decline following the introduction of ChatGPT

Replit detail in a blog post how they train Large Language Models (LLMs) using Databricks, Hugging Face, and MosaicML

EleutherAI collect important equations about transformer language models, along with related knowledge about where they come from and why they matter.

Live Science - In the early universe, gravity may have been capable of creating light, a new theoretical paper finds.

NY TIMES - The tech giant is sprinting to protect its core business with a flurry of projects, including updates to its search engine and plans for an all-new one.