Data scarcity has been a long standing issue in the field of open-domain social dialogue. To quench this thirst, we present SODA: the first publicly available, million-scale high-quality social dialogue dataset. By contextualizing social commonsense knowledge from a knowledge graph, we are able to distill an exceptionally broad spectrum of social interactions from a large language model. Human evaluation shows that conversations in SODA are more consistent, specific, and (surprisingly) natural than those in prior human-authored datasets. Using SODA, we train COSMO: a generalizable conversation model that is significantly more natural and consistent on unseen datasets than best-performing conversation models (e.g., GODEL, BlenderBot-1, Koala, Vicuna). Experiments reveal COSMO is sometimes even preferred to the original human-written gold responses. Additionally, our results shed light on the distinction between knowledge-enriched conversations and natural social chitchats. We plan to make our data, model, and code public.

We introduce Cosmos, a framework for object-centric world modeling that is designed for compositional generalization (CG), i.e., high performance on unseen input scenes obtained through the composition of known visual "atoms." The central insight behind Cosmos is the use of a novel form of neurosymbolic grounding. Specifically, the framework introduces two new tools: (i) neurosymbolic scene encodings, which represent each entity in a scene using a real vector computed using a neural encoder, as well as a vector of composable symbols describing attributes of the entity, and (ii) a neurosymbolic attention mechanism that binds these entities to learned rules of interaction. Cosmos is end-to-end differentiable; also, unlike traditional neurosymbolic methods that require representations to be manually mapped to symbols, it computes an entity's symbolic attributes using vision-language foundation models. Through an evaluation that considers two different forms of CG on an established blocks-pushing domain, we show that the framework establishes a new state-of-the-art for CG in world modeling.

It's Game Developers Conference week, which means one of two things for those working in the industry: either they're jetlagged in some hotel bar in San Francisco spending $10 a beer; or they're at home, avoiding Twitter to lessen the Fomo. GDC has long been the nexus of the games industry, a place where indie developers get their games signed and funded, coders and artists and sound designers share the techniques of their craft, and juicy development stories are overheard in the convention centre's corridors. But GDC has long been regarded by some as increasingly elitist. To its credit, it partners with plenty of different organisations to offer scholarships and sponsorships to underrepresented devs, but a regular ticket costs $1,521, and an all-access pass is $2,204. Then there's the cost of attending, which is even more prohibitive thanks to San Francisco's ever-rising prices.

Alex Wiltschko opens a black plastic suitcase and pulls out about 60 glass vials. Each contains a different scent. One smells starchy with soft floral notes, like jasmine rice cooking. Another brings to mind ocean air and the white rind of a watermelon. One is like saffron with hints of leather and black tea.

More than two years since its launch as an iOS exclusive, Osmo's kids' learning system is finally coming to Android by way of Amazon's Fire tablets. For the uninitiated, the kits use reflective AI tech, fitted to your slate's camera, to integrate physical objects into a digital environment. The more creative you get in the real-world, the bigger the digital rewards. Osmo says its games are teaching kids how to code in over 30,000 schools, and the lower-priced Amazon Fire tablets (some of which cost less than the $80-$100 kits themselves) should open it up to even more young'uns. You can pre-order both the Osmo Genius Kit and Creative Kit starting today -- both of which work on Amazon's Fire 7, Fire HD 8 and Fire HD 10 tablets -- giving you access to games like Numbers, Words, Tangrams, Newton and Masterpiece, and Monster.

Toys that teach aren't a new thing, but a growing number are calling for kids to build with blocks, circuits or everyday items before reaching for a tablet screen. Play is how kids learn about the world around them, whether it's a toddler throwing a ball or teens playing video games. It's about seeing how things work and what happens when they do something. This photo provided by Osmo shows an Osmo Creative Set. Toys that teach aren't a new thing, but a growing number are calling for kids to build with blocks, circuits or everyday items before reaching for a tablet screen. With the $63 Osmo, kids learn everything from spelling to coding not by touching a screen, but by snapping together magnetic blocks.

Jefferson Graham reviews GoPro's new Karma drone, an affordable flying camera with stellar video quality. LOS ANGELES -- Welcome to the Drone Wars, where longtime action camera manufacturer GoPro has joined No. 1 drone manufacturer DJI with its own take on a flying camera. The bad: Karma is actually heavier than we are led to believe from marketing videos, (2.2 pounds vs. 1.6 for the new $750 DJI Mavic) and some of the cool operational drone features from the Mavic just aren't there in the Karma. That said, DJI knows drones, and GoPro knows photography. The Karma trumps the Mavic with a superior camera, but Mavic is the better drone.

It's been quite the season for drone enthusiasts, with not one but two reasonably priced folding drones you can stash in a backpack arriving this month alone. Last week saw the release of the much-anticipated GoPro Karma. Not to be out-droned, market leader DJI announced the Mavic Pro, which folds up into a cute little rectangle. It's got the features you'd expect from DJI, including a 4K camera stabilized by a 3-axis gimbal and the company's handy automated flight features like collision-avoidance, TapFly, and precision hovering. You can even control this whirlybird with gestures and snap hands-free selfies as it hovers overhead. Flight time is 27 minutes, and can exceed 40 mph in "Sport" mode.