Nintendo just held a Nintendo Direct. No, not the Switch 2 Direct you've been waiting for, but another one. In what will potentially be the last Switch 1-focused Direct livestream, Nintendo made a bundle of announcements, some of which weren't even games. Instead of wasting time, let's get right to it. Last year, Square Enix released a really gorgeous "HD-2D" remake of Dragon Quest III, with the promise that the first two games in the series would eventually get the same treatment.

Large organizations wanting to integrate generative AI into their business can find the process daunting, especially if they're unsure how or where to start. A new initiative with Anthropic's Claude AI aims to provide the tools for enterprises struggling with this challenge. In a new five-year partnership announced Thursday, Anthropic and AI company Databricks will offer Claude AI models through the Databricks Data Intelligence Platform, which helps businesses use AI with their internal data. The deal will bring the tools and technology to more than 10,000 companies. Anthropic's latest Claude 3.7 Sonnet model is already accessible through Databricks on AWS, Azure, and Google Cloud. This model uses advanced reasoning and greater processor time to evaluate your request step-by-step and then produce a detailed result.

It might be around that time of year when you're starting to figure out your summer vacation plans. Google has revealed some new features that can help with that, including a handy AI-powered one for Maps. If you turn on the new screenshot list, Gemini can automatically recognize locations that are mentioned in screenshots you take in the app. You can then save the places you're interested in to a list. These saved spots will appear on the map, and you can share the list with your travel companions.

Normally priced at 275, this is the best discount we've seen on the 2-in-1 robot, matching the sale price we saw during the holiday shopping season last year. As you can see in our roundup of the best budget robot vacuums, we've consistently rated iRobot's machines highly for their reliability and ease of use. The Roomba Combo Essential is a simple, no-frills option that both vacuums and mops, making it a solid pick for small apartments, dorm rooms or anyone who wants a cleaner floor without spending a fortune. That's more than 100 off this robot that can vacuum and mop. The vacuum uses special multi-surface brushes to pick up dirt, dust and pet hair from hard floors and carpets.

It's no secret that large language models work in mysterious ways. Few--if any--mass-market technologies have ever been so little understood. That makes figuring out what makes them tick one of the biggest open challenges in science. Shedding some light on how these models work would expose their weaknesses, revealing why they make stuff up and can be tricked into going off the rails. It would help resolve deep disputes about exactly what these models can and can't do.

The scientists didn't have high expectations when they asked their AI model to complete the poem. "He saw a carrot and had to grab it," they prompted the model. "His hunger was like a starving rabbit," it replied. The rhyming couplet wasn't going to win any poetry awards. But when the scientists at AI company Anthropic inspected the records of the model's neural network, they were surprised by what they found.

Recent advancements in Multimodal Large Language Models (MLLMs) have greatly improved their abilities in image understanding. However, these models often struggle with grasping pixel-level semantic details, e.g., the keypoints of an object. To bridge this gap, we introduce the novel challenge of Semantic Keypoint Comprehension, which aims to comprehend keypoints across different task scenarios, including keypoint semantic understanding, visual prompt-based keypoint detection, and textual prompt-based keypoint detection. Moreover, we introduce KptLLM, a unified multimodal model that utilizes an identify-then-detect strategy to effectively address these challenges. KptLLM underscores the initial discernment of semantics in keypoints, followed by the precise determination of their positions through a chain-of-thought process. With several carefully designed modules, KptLLM adeptly handles various modality inputs, facilitating the interpretation of both semantic contents and keypoint locations. Our extensive experiments demonstrate KptLLM's superiority in various keypoint detection benchmarks and its unique semantic capabilities in interpreting keypoints.

This paper investigates simple bilevel optimization problems where we minimize an upper-level objective over the optimal solution set of a convex lower-level objective. Existing methods for such problems either only guarantee asymptotic convergence, have slow sublinear rates, or require strong assumptions. To address these challenges, we propose a penalization framework that delineates the relationship between approximate solutions of the original problem and its reformulated counterparts.