Long-context modelling for large language models (LLMs) has been a key area of recent research because many real world use cases require reasoning over longer inputs such as documents. The focus of research into modelling long context has been on how to model position and there has been little investigation into other important aspects of language modelling such as instruction tuning. Long context training examples are challenging and expensive to create and use. In this paper, we investigate how to design instruction data for the post-training phase of a long context pre-trained model: how much and what type of context is needed for optimal and efficient post-training. Our controlled study reveals that models instruction-tuned on short contexts can effectively generalize to longer ones, while also identifying other critical factors such as instruction difficulty and context composition. Based on these findings, we propose context synthesis, a novel data synthesis framework that leverages off-the-shelf LLMs to generate extended background contexts for high-quality instruction-answer pairs. Experiment results on the document-level benchmark (LongBench) demonstrate that our proposed approach outperforms previous instruction synthesis approaches and comes close to the performance of human-annotated long-context instruction data. The project will be available at: https://github.com/NJUNLP/context-synthesis.

We address an important gap in detecting political bias in news articles. Previous works that perform document classification can be influenced by the writing style of each news outlet, leading to overfitting and limited generalizability. Our approach overcomes this limitation by considering both the sentence-level semantics and the document-level rhetorical structure, resulting in a more robust and style-agnostic approach to detecting political bias in news articles. We introduce a novel multi-head hierarchical attention model that effectively encodes the structure of long documents through a diverse ensemble of attention heads. While journalism follows a formalized rhetorical structure, the writing style may vary by news outlet. We demonstrate that our method overcomes this domain dependency and outperforms previous approaches for robustness and accuracy. Further analysis and human evaluation demonstrate the ability of our model to capture common discourse structures in journalism. Our code is available at: https://github.com/xfactlab/emnlp2023-Document-Hierarchy

'Gutfeld!' panelists discuss whether every generation has become less attractive. We'll get to the indictments and the Bidens in the next block. GUTFELD MUSIC VIDEO: So many stories too upsetting to look into. A bunch of news that just makes you wanna cry. Then suddenly everything gets less depressing.

There are specific use cases like the spam filter, where doing traditional programming is hard. Also, the real use of machine learning, that is, cognitive problems, such as image recognition, speech processing, Natural Language Processing (NLP), and so on. These tasks are extremely data-driven and complex, and solving them using rules would be a nightmare. So, an increase in complexity and data-driven problems are the key areas where machine learning can thrive. For example, we have NLP models that can write entire movie scripts, image processing models that can colorize old black and white images, and so on.

"We live in a network world. Everything we do is an outcome of multiple elements. The pervasion of social media in our lives means hundreds and thousands of tweets and retweets by the minute. Gone are the times when information asymmetry was exploited," remarked Dr Alok Chaturvedi, professor of Management and Computer Science, Purdue University while initiating a talk at ORF Delhi on Big Data and Agent Based Simulation for Policy Analysis on 8 May, 2018. The discussion was moderated by Rakesh Sood, Distinguished Fellow, ORF and a former ambassador.

Walmart's shelf-scanning robots have started picking up shifts in some California stores. The six-foot-tall robots come with a tower that tower that is fitted with cameras that scan aisles to check stock and identify missing and misplaced items, as well as incorrect prices and mislabeling. The robots pass that data to store employees, who then stock the shelves and fix errors. A Walmart in Milpitas, Calif. is currently testing the robot, but Walmart also has plans to roll them out to 50 more U.S. stores sometime soon. The approximately 2-foot (0.61-meter) robots come with a tower that is fitted with cameras that scan aisles to check stock and identify missing and misplaced items, incorrect prices and mislabeling Out-of-stock items are a big problem for retailers since they miss out on sales every time a shopper cannot find a product on store shelves.

Grimes has spent her career at Google, where she currently works on data-driven resource planning, cost analysis, and distributed cluster management software as part of the Technical Institute Group. Grimes holds a PhD in Statistics from Stanford University and an AB in Anthropology from Harvard University. Meta S. Brown is a consultant, speaker and writer who promotes the use of business analytics. A hands-on analyst who has tackled projects with up to $900 million at stake, she is a recognized expert in cutting-edge business analytics. Jennifer Chayes is a Distinguished Scientist and Managing Director at Microsoft Research.

Wal-Mart's chief technology officer, Jeremy King, points out that this is a task humans aren't very efficient at doing. "If you are running up and down the aisle and you want to decide if we are out of Cheerios or not, a human doesn't do that job very well, and they don't like it," he said. The robots are 50% more productive at scanning shelves and can complete the task three times faster than employees, while being much more accurate. Currently, associates can only scan store shelves twice per week.

Walking up and down aisles of a retail store, scanning row after row of products to help keep track of inventory is one of the most tedious and time-consuming tasks store employees endure, and most dread the assignment. This is especially true for a company like Wal-Mart Stores (NYSE:WMT), which can have as many as 200,000 products lining its shelves. The cost of not completing this monotonous chore can be high. Every out-of-stock item can result in a lost sale for the retailer. This is particularly critical as brick-and-mortar stores contend with the rise of online shopping from the likes of Amazon.com

Information-technology executives at companies of all kinds in the Drucker Institute's Management Top 250 ranking of the most effectively managed U.S. companies have found ways to incorporate the best strategies from the technology sector to promote innovation, collaboration and new business models. IT executives at companies such as Wal-Mart Stores Inc., WMT 0.01% Procter & Gamble PG -1.26% Co.--both among the 20 companies with the highest scores for innovation--and Capital One Financial Corp. COF 1.28% are taking on a more central role in the business, helping their companies adapt to the digital age. To that end, they're using cloud services and collaborative work models to speed up the development and delivery of technology, automating mundane work processes for employees and embracing cutting-edge technologies that add business value. "We're modeling ourselves off the best technology companies out there," says Rob Alexander, Capital One's chief information officer. "Not legacy tech companies, but companies that have been built in the era of the cloud and the internet."