Large language models (LLMs) adapted to follow user instructions are now widely deployed as conversational agents. In this work, we examine one increasingly common instruction-following task: providing writing assistance to compose a long-form answer. To evaluate the capabilities of current LLMs on this task, we construct KIWI, a dataset of knowledge-intensive writing instructions in the scientific domain. Given a research question, an initial model-generated answer and a set of relevant papers, an expert annotator iteratively issues instructions for the model to revise and improve its answer. We collect 1,260 interaction turns from 234 interaction sessions with three state-of-the-art LLMs. Each turn includes a user instruction, a model response, and a human evaluation of the model response. Through a detailed analysis of the collected responses, we find that all models struggle to incorporate new information into an existing answer, and to perform precise and unambiguous edits. Further, we find that models struggle to judge whether their outputs successfully followed user instructions, with accuracy at least 10 points short of human agreement. Our findings indicate that KIWI will be a valuable resource to measure progress and improve LLMs' instruction-following capabilities for knowledge intensive writing tasks.

Start-up Eddy Travels has launched a new artificial intelligence group planning digital assistant as part of a partnership with flight search specialist Kiwi.com. Tequila API powers the Eddy Travels AI assistant flight search which was launched in 2018 and has four million users. Eddy Travels, the co-located Lithuanian and Canadian firm, said the new group chat planning service will help families and friends planning to reconnect after months in lockdown. Edmundas Balčikonis, co-founder and chief executive of Eddy Travels, said: "The new chat app removes the burden for one person to be responsible for planning everything and makes it a fun shared activity, even if it's done remotely for now." Oliver Dlouhý, co-founder and chief executive of Kiwi.com, added: "Edmundas and his team did an amazing job building the first group travel booking app that actually works and is really fun to use. "We are proud to be providers of flight and ground transportation inventory, including Virtual Interlining.

Kiwi announced today that its rover robots are now delivering food and other items around two areas in San Jose, CA. According to Kiwi, what makes this rollout unique is that the company has partnered with the city itself to help mitigate deployment and maintenance issues that come with robots scurrying around city sidewalks. Starting today, a "squad" of 25 robots will be making deliveries in the downtown area and Buena Vista neighborhoods of San Jose. As many as 120 businesses will be participating in the program. Kiwi is also working with the city of San Jose to help with the management of the roaming robots.

Scientists who designed an artificially intelligent robot that helped children with autism boost their learning and social skills hope such technology could one day aid others with the developmental disorder. The study saw seven children with mild to moderate autism take home what is known as a socially assistive robot, named Kiwi, for a month. According to a statement by the University of Southern California where the team is based, the participants from the Los Angeles area were aged between three and seven years old, and played space-themed games with the robot almost daily. As Kiwi was fitted with machine-learning technology, it was able to provide unique feedback and instructions to the children based on their abilities. For instance, if the child got a question wrong Kiwi would give prompts to help them solve it, and tweak the difficulty levels to challenge the child appropriately.

Researchers have developed a new personalized learning robot for autistic children that uses machine learning to adapt its lessons to each kid's changing needs. The University of Southern California team put a "socially assistive robot" called Kiwi in the homes of 17 autistic children and set the two-foot-tall, green-feathered robot to give each child personalized classes. Over the course of a month, the children played space-themed math games on a tablet device while Kiwi provided feedback and instruction, such as congratulating them on a correct answer or giving tips after a wrong one. As the lessons progressed, algorithms adjusted Kiwi's feedback and the difficulty of the games to the child's individual needs. By the end of the month, all of the children had improved their math skills, while 92% had also improved their social skills.

Researchers have whittled down a massive 8 million tweets, to a more manageable 1.2 million to look at how te reo MÄ ori is being used in the genre. The team from the University of Waikato have focused on 77 MÄ ori loanwords (te reo MÄ ori words used in an English context) and used them as training data for their machine-learning model. Machine learning allows data scientists to provide a computer with a large data set, and teach it to make predictions based on that data. Computing and Mathematical Sciences student David Trye spent the summer working on the project, with supervisorsDr Andreea Calude and Dr Felipe Bravo Márquez. The initial 8-million tweets contained a fair bit of distracting data'noise'.

A roving delivery robot made the mistake of dropping off a student's takeout order well-done after it caught fire at the University of California, Berkeley. On Friday, students found a Kiwi delivery robot consumed by flames in the middle of campus. A 30-second clip of the incident shows people looking on, many with their phones pulled out, watching idly as the four-wheeled droid burns, until a passerby finally puts out the flames with a fire extinguisher. Kiwi Campus, the startup behind the device, later explained that the mishap was due to'human error' and that it was removing all of its other robots from operation until it resolved an issue with the device's defective battery. The delivery robots are made by Kiwi, a California-based startup that works out of University of California, Berkeley's startup incubator. Each delivery robot is a tiny, four-wheeled device that's about the size of a small dog.

A food delivery robot was destroyed after it caught fire because of "human error", its creator has confirmed. Kiwibot autonomous delivery robots have been rolling around the University of California, Berkeley campus for two years. On Friday, students found one of the robots in flames and shared photos on social media. Kiwi said the cause was a "defective battery" that had been accidentally installed in the robot. "One of the batteries for our robot that was idling started smouldering, eventually leading to some smoke and minor flames," the company said in a statement.

When asked what kind of company it is, the team at Kiwi Campus always says that it is a delivery company, not a robotics firm -- even though in these early days, its delivery robots are probably the most recognizable thing about the company. But, according to co-founder and CTO Jason Oviedo, when he and co-founder and CEO Felipe Chavez Cortes were first thinking about founding a firm, robotics wasn't on their minds. Building a delivery service -- for students, by students -- was. However, Oviedo told PYMNTS in a recent interview, it didn't take them very long to figure out that pure play delivery wasn't a market where they could make a lot of progress, and it wasn't until they started revising that initial idea that they started thinking about automation -- and ultimately delivery robots -- as their best entry point for the market. "That makes a huge different because our entire process and design isn't centered on building the flashiest robot -- it is about how to use a robot to make the delivery process faster, cheaper and better."

TACAIR-SOAR flew all U.S. fixed-wing aircraft. The general goal was to generate behavior that "looks human" when viewed by a training audience participating in operational military exercises. Its most dramatic use was in the Synthetic Theater of War 1997 (STOW '97), held 29-31 October 1997 (Ceranowicz, 1998; Laird, Jones, and Nielsen 1998; Laird et al. 1998). STOW '97 was a United States Department of Defense (DoD) Advanced Concept Technology Demonstration (ACTD) that was integrated with the United Endeavor 98-1 (UE 98-1) training exercise. As an ACTD, the overall goal of STOW '97 was to permit an early and inexpensive evaluation of advanced technologies that show promise for improving military effectiveness.