Breakthroughs, discoveries, and DIY tips sent every weekday. When classic films undergo 4K restorations, the results can divide fans. Look around Hollywood and you'll find numerous examples of movie rereleases featuring controversial uses of digital noise reduction, motion smoothing, and other post-production tools. Meanwhile, the proliferation of AI- and machine learning-based upscaling programs has only complicated the debate. When approached properly, though, the technique has helped revive some of Hollywood's oldest--and for a long time, inaccessible--movies.

The data-to-equation (Data2Eqn) task aims to discover interpretable mathematical equations that map observed values to labels, offering physical insights and broad applicability across academic and industrial domains. Genetic programming and traditional deep learning-based approaches suffer from search inefficiency and poor generalization on small task-specific datasets. Foundation models showed promise in this area, but existing approaches suffer from: 1) They are pretrained on general-purpose data distributions, making them less effective for domain-specific tasks; and 2) their training objectives focus on token-level alignment, overlooking mathematical semantics, which can lead to inaccurate equations. To address these issues, we aim to enhance the domain adaptability of foundation models for Data2Eqn tasks. In this work, we propose a reinforcement learning-based finetuning framework that directly optimizes the generation policy of a pretrained model through reward signals derived from downstream numerical fitness. Our method allows the model to adapt to specific and complex data distributions and generate mathematically meaningful equations. Extensive experiments demonstrate that our approach improves both the accuracy and robustness of equation generation under complex distributions.

If you've performed a Google search lately, you've undoubtedly come across an AI Overview in your search results. This tool, powered by Google's Gemini, tries to save you some clicks by aggregating information from the links populated in your search results and succinctly delivering what it believes to be the information you're looking for. The accuracy of these overviews, however, often leaves a lot to be desired, and the tool has been plagued with hallucinations since its launch (with varying degrees of hilarity). Now Google is bringing the tool to YouTube, testing a video version of AI overviews for a small number of YouTube Premium members in the US across limited English search queries. While Google search results show LLM-generated text summaries, YouTube's AI overviews will function as something of a highlight reel for certain videos.

Moving your body is for everyone, regardless of experience level, skill, or location. This year's Best of What's New innovations make getting outside and active easier in many ways. A tightly woven shirt stops itchy mosquito bites sans chemicals. An electric fishing reel cuts the cord and ditches heavy batteries once and for all. An app combines avalanche education with hard-to-find reports for safer snowshoeing and skiing.

Meta just announced an intriguing tool that uses AI to automatically dub Reels into other languages, complete with lip-sync. This feature was revealed at the annual Meta Connect livestream event and was introduced by CEO Mark Zuckerberg. Zuckerberg showed this off during the keynote, and everything seemed to work flawlessly. The technology not only translates the content, according to Meta, but will also "simulate the speaker's voice in another language and sync their lips to match." It's worth noting, however, that this didn't appear to be a live demo, but was still pretty impressive.

Meta is revamping how Facebook recommends videos across Reels, Groups, and the main Facebook Feed, by using AI to power its video recommendation algorithm, Facebook head Tom Alison revealed on Wednesday. The world's largest social network has already switched Reels, its TikTok competitor, to the new engine, and plans to use it in all places within Facebook that show video -- the main Facebook feed and Groups -- as part of a "technology roadmap" through 2026, Alison said at a Morgan Stanley tech conference in San Francisco. Meta has made competing with TikTok a top priority ever since the app, which serves up vertical video clips and is known for its powerful recommendation engine that seems to know exactly what will keep users hooked, started exploding in popularity in the US in the last few years. When Facebook tested the new AI-powered recommendation engine with Reels, watch time went up by roughly 8 to 10 percent, Alison revealed. "So what that told us was this new model architecture is learning from the data much more efficiently than the previous generation," Alison said. "So that was like a good sign that says, OK, we're on the right track."

Adobe just released the latest iterations of Photoshop Elements and Premiere Elements. These 2024-branded versions feature plenty of new features that streamline the creative process, many of them aided by, wait for it, artificial intelligence. Beyond AI-powered tools, there's also some other stuff for photo and video editors to get excited about. Let's start with AI features, all of which are powered by Adobe's new Sensei AI platform. On the Photoshop side of things, there's a new tool that automatically selects objects and backgrounds for removal, editing or replacement.

Short videos on social media are the dominant way young people consume content. News outlets would like to reach audiences through news reels - short videos that convey news - but struggle to translate traditional journalistic formats into short, colloquial videos. Generative AI has the potential to transform content but often fails to be correct and coherent by itself. To help journalists create scripts and storyboards for news reels, we introduce a human-AI co-creative system called ReelFramer. It uses an intermediate step of framing and foundation to guide AI toward better outputs. We introduce three narrative framings to balance information and entertainment in news reels. The foundation for the script is a premise, and the foundation for the storyboard is a character board. Our studies show that the premise helps generate more relevant and coherent scripts and that co-creating with AI lowers journalists' barriers to making their first news reels.

Atari is launching another retro home console, after its last effort. The Atari 2600 pays homage to the original Atari 2600, launched in 1977, but this remake echoes the four-switch model from 1980. The console has been "lovingly recreated to the same specifications as the original" but is only 80 percent of its size. The console's plus features are the HDMI output and widescreen support. It'll have 10 titles in the box, but Atari die-hards will want to track down physical cartridges if they want to play the big hits of the era, like Pac-Man or Pitfall!

Despite their growing popularity, many public kiosks with touchscreens are inaccessible to blind people. Toucha11y is a working prototype that allows blind users to use existing inaccessible touchscreen kiosks independently and with little effort. Toucha11y consists of a mechanical bot that can be instrumented to an arbitrary touchscreen kiosk by a blind user and a companion app on their smartphone. The bot, once attached to a touchscreen, will recognize its content, retrieve the corresponding information from a database, and render it on the user's smartphone. As a result, a blind person can use the smartphone's built-in accessibility features to access content and make selections. The mechanical bot will detect and activate the corresponding touchscreen interface. We present the system design of Toucha11y along with a series of technical evaluations. Through a user study, we found out that Toucha11y could help blind users operate inaccessible touchscreen devices.