Retrieval-augmented generation (RAG) faces challenges related to factual correctness, source attribution, and response completeness. To address them, we propose a modular pipeline for grounded response generation that operates on information nuggets-minimal, atomic units of relevant information extracted from retrieved documents. The multistage pipeline encompasses nugget detection, clustering, ranking, top cluster summarization, and fluency enhancement. It guarantees grounding in specific facts, facilitates source attribution, and ensures maximum information inclusion within length constraints. Extensive experiments on the TREC RAG'24 dataset evaluated with the AutoNuggetizer framework demonstrate that GINGER achieves state-of-the-art performance on this benchmark.

People can acquire knowledge in an unsupervised manner by reading, and compose the knowledge to make novel combinations. In this paper, we investigate whether pretrained language models can perform compositional generalization in a realistic setting: recipe generation. We design the counterfactual recipe generation task, which asks models to modify a base recipe according to the change of an ingredient. This task requires compositional generalization at two levels: the surface level of incorporating the new ingredient into the base recipe, and the deeper level of adjusting actions related to the changing ingredient. We collect a large-scale recipe dataset in Chinese for models to learn culinary knowledge, and a subset of action-level fine-grained annotations for evaluation. We finetune pretrained language models on the recipe corpus, and use unsupervised counterfactual generation methods to generate modified recipes. Results show that existing models have difficulties in modifying the ingredients while preserving the original text style, and often miss actions that need to be adjusted. Although pretrained language models can generate fluent recipe texts, they fail to truly learn and use the culinary knowledge in a compositional way. Code and data are available at https://github.com/xxxiaol/counterfactual-recipe-generation.

San Francisco, California-based Grammarly, which develops an AI-powered writing assistant, today announced that it raised $200 million, valuing the company at $13 billion post-money. Baillie Gifford led the round with participation from funds and accounts managed by BlackRock. Grammarly CEO Brad Hoover says the funding will be put toward further developing the company's technology and "accelerat[ing] efforts to help people communicate … in our digital-first world." As enterprises increasingly embrace digitization, the over $1.2 billion AI writing assistant market is expected to grow at a compound annual growth rate of 27.6% from 2018 to 2028. According to a survey from John Snow Labs and Gradient Flow, 60% of tech leaders indicated that their budgets for natural language processing -- which encompasses technologies like Grammarly's -- grew by at least 10% compared to 2020, while a third said that their spending climbed by more than 30%.

Chinese AI businesses have been growing rapidly since 2010. They have attracted significant investment from Internet giants and a vast number of emerging AI companies have emerged. Over the past decade, Chinese AI start-ups have gradually moved away from noisy bubbles and landed in an investment boom. In 2020, when people were fighting against the pandemic, CloudMinds, an AI start-up based in Beijing, developed a humanoid service robot named Cloud Ginger XR-1. Ginger played an important role in local hospitals, delivering food and medication to patients in a contactless manner when it was needed the most. Moreover, Ginger entertained patients, freeing up doctors and medical teams to focus on more critical health matters.

For entrepreneurial MIT students looking to put their skills to work for a greater good, the Media Arts and Sciences class MAS.664 (AI for Impact) has been a destination point. With the onset of the pandemic, that goal came into even sharper focus. Just weeks before the campus shut down in 2020, a team of students from the class launched a project that would make significant strides toward an open-source platform to identify coronavirus exposures without compromising personal privacy. Their work was at the heart of Safe Paths, one of the earliest contact tracing apps in the United States. The students joined with volunteers from other universities, medical centers, and companies to publish their code, alongside a well-received white paper describing the privacy-preserving, decentralized protocol, all while working with organizations wishing to launch the app within their communities.

Identifying how much personalization to offer – and to whom – will separate winners from losers. Hyper-personalization is one of three areas we focus on as part of the digital consumption cross-industry theme. The other themes we examine are products and services to experiences and ownership to access. More than 70% of customers now expect more personalized experiences with the brands they interact with,¹ and digital technology is enabling companies to meet these expectations by delivering personalization to large numbers of customers at a low cost. Spectacular advances in artificial intelligence (AI) and software intelligence are enabling companies to take personalization to the next level, making products and services highly relevant to a very large number of customers at the same time.

As the Brett Kavanaugh hearings dominated the news cycle in September, Silicon Valley–based mental health startup Ginger found its app buzzing with sexual assault survivors who were reporting feelings of heightened anxiety, anger and powerlessness. It scanned the words users typed to their therapists in a bid to better understand the patient's situation and then recommended how the health professionals might intervene. The therapists were then able to provide coping strategies based on an individual's needs. For Ginger co-founder Karan Singh, the reason for developing the app was personal. After learning of a friend's suicide attempt, Singh decided to help develop better resources for people suffering from depression.

The United States faces a mental health epidemic. Nearly one in five American adults suffers from a form of mental illness. Suicide rates are at an all-time high, 115 people die daily from opioid abuse, and one in eight Americans over 12 years' old take an antidepressant every day. The economic burden of depression alone is estimated to be at least $210 billion annually, with more than half of that cost coming from increased absenteeism and reduced productivity in the workplace. In a crisis that has become progressively dire over the past decade, digital solutions -- many with artificial intelligence (AI) at their core -- offer hope for reversing the decline in our mental wellness.

"How are you doing today?" "What's going on in your world right now?" "How do you feel?" These might seem like simple questions a caring friend would ask. However, in the present day of mental health care, they can also be the start of a conversation with your virtual therapist. Advancements in artificial intelligence (AI) are bringing psychotherapy to more people who need it. It is becoming clear that AI for mental health could be a game changer.

Robots could soon lend nurses a helping hand in the delivery ward. In a two-year study, researchers from MIT have investigated whether robots can effectively take on the job of a'resource nurse,' making complex decisions in a fast-paced environment. Resource nurses are tasked with making thousands of critical decisions, including bed assignments and selecting the right nurse to perform a C-section – and so far, the researchers found that doctors and nurses accepted the robot-made recommendations 90 percent of the time. Researchers from MIT's Computer Science and Artificial Intelligence Laboratory trained a Nao robot to learn the differences between good and bad decisions on the delivery floor at Beth Israel Deaconess Medical center in Boston In the study, the researchers from MIT's Computer Science and Artificial Intelligence Laboratory trained a Nao robot to learn the differences between good and bad decisions on the delivery floor, according to CNN Money. Then, the robot named'Ginger' was taken to Beth Israel Deaconess Medical Center in Boston to test out her skills.