Understanding and interpreting the internal representations of large language models (LLMs) remains an open challenge. Patchscopes introduced a method for probing internal activations by patching them into new prompts, prompting models to self-explain their hidden representations. We introduce Superscopes, a technique that systematically amplifies superposed features in MLP outputs (multilayer perceptron) and hidden states before patching them into new contexts. Inspired by the "features as directions" perspective and the Classifier-Free Guidance (CFG) approach from diffusion models, Superscopes amplifies weak but meaningful features, enabling the interpretation of internal representations that previous methods failed to explain-all without requiring additional training. This approach provides new insights into how LLMs build context and represent complex concepts, further advancing mechanistic interpretability.

Sampling from diffusion models involves a slow iterative process that hinders their practical deployment, especially for interactive applications. To accelerate generation speed, recent approaches distill a multi-step diffusion model into a single-step student generator via variational score distillation, which matches the distribution of samples generated by the student to the teacher's distribution. However, these approaches use the reverse Kullback-Leibler (KL) divergence for distribution matching which is known to be mode seeking. In this paper, we generalize the distribution matching approach using a novel $f$-divergence minimization framework, termed $f$-distill, that covers different divergences with different trade-offs in terms of mode coverage and training variance. We derive the gradient of the $f$-divergence between the teacher and student distributions and show that it is expressed as the product of their score differences and a weighting function determined by their density ratio. This weighting function naturally emphasizes samples with higher density in the teacher distribution, when using a less mode-seeking divergence. We observe that the popular variational score distillation approach using the reverse-KL divergence is a special case within our framework. Empirically, we demonstrate that alternative $f$-divergences, such as forward-KL and Jensen-Shannon divergences, outperform the current best variational score distillation methods across image generation tasks. In particular, when using Jensen-Shannon divergence, $f$-distill achieves current state-of-the-art one-step generation performance on ImageNet64 and zero-shot text-to-image generation on MS-COCO. Project page: https://research.nvidia.com/labs/genair/f-distill

TARGETING WASTE: Albert Invent CEO Nick Talken shared how his artificial intelligence platform saves thousands of scientists time and money on "Mornings with Maria," saying government can also benefit from the technology. GRAB A'BYTE': It was a big week for Yum Brands' Taco Bell as executives from the fast-food giant held its annual Live Más LIVE event in New York City, showcased new labor-saving technology, and announced an investment of 1 billion into digital and technology. AVOID IRS SCAMS: Tax season is upon us, and while many of you are preparing to file your returns, it's crucial to be aware of the ever-evolving world of tax scams. Scam written on tax forms (Kurt "CyberGuy" Knutsson) GOLDEN VOICE: A producer for the Oscar-winning film, "The Brutalist," is defending the production's use of artificial intelligence. D.J. Gugenheim at the 97th Oscars held at the Dolby Theatre on March 02, 2025 in Hollywood, California.

A warrior is in a prison cell. His guard approaches and shows him the wooden sword that he will receive once he has earned his freedom. The warrior grabs it, uses his unlocked cell door to knock the guard down, and places the sword's tip on the guard's throat. He drives it in as one might hammer a post, a coarse and grisly death. Then, for some reason, swaying back and forth, the warrior yells down at the corpse, "Wood or steel, a point is still a point!"

Among robots, a front flip is significantly more difficult than a backflip. Chinese robotics company Zhongqing Robotics, also known as EngineAI, has officially entered the humanoid robotics scene by releasing a video showcasing what it claims is the world's first humanoid robot front flip. Robot backflips are becoming commonplace, but a front flip is significantly more difficult than a backflip, as any gymnast can attest. GET EXPERT SECURITY ALERTS, MUST-KNOW TECH TIPS, AND THE LATEST DIGITAL TRENDS -- STRAIGHT TO YOUR INBOX. Unlike humans, robots rely on precise sensor data and motor control to execute complex movements.

Conversational Recommender Systems (CRSs) have emerged as a transformative paradigm for offering personalized recommendations through natural language dialogue. However, they face challenges with knowledge sparsity, as users often provide brief, incomplete preference statements. While recent methods have integrated external knowledge sources to mitigate this, they still struggle with semantic understanding and complex preference reasoning. Recent Large Language Models (LLMs) demonstrate promising capabilities in natural language understanding and reasoning, showing significant potential for CRSs. Nevertheless, due to the lack of domain knowledge, existing LLM-based CRSs either produce hallucinated recommendations or demand expensive domain-specific training, which largely limits their applicability. In this work, we present G-CRS (Graph Retrieval-Augmented Large Language Model for Conversational Recommender Systems), a novel training-free framework that combines graph retrieval-augmented generation and in-context learning to enhance LLMs' recommendation capabilities. Specifically, G-CRS employs a two-stage retrieve-and-recommend architecture, where a GNN-based graph reasoner first identifies candidate items, followed by Personalized PageRank exploration to jointly discover potential items and similar user interactions. These retrieved contexts are then transformed into structured prompts for LLM reasoning, enabling contextually grounded recommendations without task-specific training. Extensive experiments on two public datasets show that G-CRS achieves superior recommendation performance compared to existing methods without requiring task-specific training.

Egocentric human body estimation allows for the inference of user body pose and shape from a wearable camera's first-person perspective. Although research has used pose estimation techniques to overcome self-occlusions and image distortions caused by head-mounted fisheye images, similar advances in 3D human mesh recovery (HMR) techniques have been limited. We introduce Fish2Mesh, a fisheye-aware transformer-based model designed for 3D egocentric human mesh recovery. We propose an egocentric position embedding block to generate an ego-specific position table for the Swin Transformer to reduce fisheye image distortion. Our model utilizes multi-task heads for SMPL parametric regression and camera translations, estimating 3D and 2D joints as auxiliary loss to support model training. To address the scarcity of egocentric camera data, we create a training dataset by employing the pre-trained 4D-Human model and third-person cameras for weak supervision. Our experiments demonstrate that Fish2Mesh outperforms previous state-of-the-art 3D HMR models.

Whether it's a cheeky wink or a grinning cowboy, it might seem like there is already an emoji for every possible situation. But amazingly, there are now even more designs on the way. Apple has revealed eight new emojis that are officially coming to iPhones as part of the iOS 18.4 update. While they are only available in'beta' for now, these new symbols should be rolling out to everyone when the full update is released in late March or early April. In the update, users will be able to send a colourful fingerprint, bright purple splat, or the flag of the island of Sark.

After flooding our homes with every Sonos model you can buy (and filling all remaining space with the boxes of said speakers), then using them for a couple of years, we've come to value their audio fidelity and ability to network seamlessly together. There isn't another speaker system that lets you string together multiple speakers as easily or connect them to stream in different rooms of your home while keeping the audio perfectly in sync. The closest thing may be Google Assistant speakers, and Sonos connects to that system as well. Easy streaming: The Sonos app supports almost every streaming service in existence, and many apps, like Spotify, let you stream to Sonos speakers within them. The Sonos ecosystem can also handle home-theater applications and can support a full surround-sound setup.

'Beetlejuice Beetlejuice' star Justin Theroux tells Fox News Digital his thoughts on artificial intelligence and how it will impact future Hollywood films. A producer for the Oscar-winning film, "The Brutalist," is defending the production's use of artificial intelligence. D.J. Gugenheim, one of several producers involved in the film, spoke with Deadline at the Oscars on Sunday night, saying the technology is simply a tool. "If you're in post [-production] on a film, there's so many tools that you use, whether it's lighting, sound, and these are all versions of functions of numbers," he told the outlet. "What's important about how we're making a film is that we're trusting the actors and the creatives and the talent to make a film. So, if no one is losing a job, and you're making the best version of the product, that's when you're using a tool."