Fireworks by Grucci has been in the pyrotechnics business for six generations. For America's 250th, Grucci fired dozens of shows, including one that covered the entire three-and-a-half-mile Vegas Strip. A few days before the Fourth of July, I was on the rooftop of a casino in Las Vegas, where I ran into the head electrician of the Eisenhower Theatre at the John F. Kennedy Memorial Center for the Performing Arts, which is in Washington, D.C. The electrician, Thomas Benya, who is also a licensed pyrotechnician, had come to Vegas with a swarm of other pyrotechnicians, to set up, and set off, what was being billed as the nation's largest fireworks show. Back in D.C., President Donald Trump was claiming that show, around the National Mall, would be bigger, and that he would be giving a "really long" speech. Trump had been trying to muscle his name onto the Kennedy Center, and then, amid a series of staff departures and cancellations by performers, closed the place down for a "comprehensive revitalization project"; Benya, who enjoys "building art with people," and who has worked at the Kennedy Center for twenty years, since he was eighteen, told me, "It's been traumatic." Vegas seemed unlikely to offer respite. Three hundred and twenty-nine thousand visitors were in town, to celebrate America's two-hundred-and-fiftieth birthday, and all of them were on my elevator. It was a million degrees outside. On a different rooftop, on another day, a gloved woman whose T-shirt read " " paused while loading orange-strobe fireworks shells into three-inch mortars and told me, "When you've been around one hundred and twenty and above, this doesn't even feel like anything." The best sweaty person I saw in Vegas was an older woman with upswept black hair and bangs, wearing a lime-green fake chrysanthemum as plump as a grapefruit tucked behind her right ear, and carrying a Spanish fan. This was at a coffee shop in the Arts District, miles from the Strip. At the next table was a woman who also had black bangs, plus lip piercings and a tattoo on both wrists. She was engrossed in a book called "Fool's Errand," a fantasy novel about an assassin named FitzChivalry Farseer who, per a quick Google, has abandoned "self-imposed exile to rescue a missing prince and navigate perilous political waters." The woman wore an "Invader Zim" T-shirt: " " Belowground was the Vegas Loop--Elon Musk's idea.

Waymo passengers were terrified as a driverless car drove into lit fireworks on a San Francisco street during Fourth of July celebrations. No injuries were reported.

Follow this section to personalize your feed and get instant alerts. Follow Go to your personalized feed WHY FOLLOW? Smart Alerts: Get notified about major news as it happens. Follow this tag to personalize your feed and get instant alerts. Follow Go to your personalized feed WHY FOLLOW?

Spoken Language Models (SLMs) are designed to take speech inputs and produce spoken responses. However, current SLMs lack the ability to perform an internal, unspoken thinking process before responding. In contrast, humans typically engage in complex mental reasoning internally, enabling them to communicate ideas clearly and concisely. Thus, integrating an unspoken thought process into SLMs is highly desirable. While naively generating a complete chain-of-thought (CoT) reasoning before starting to talk can enable thinking for SLMs, this induces additional latency for the speech response, as the CoT reasoning can be arbitrarily long. To solve this issue, we propose Stitch, a novel generation method that alternates between the generation of unspoken reasoning chunks and spoken response chunks. Since the audio duration of a chunk of spoken response is much longer than the time to generate the tokens in a chunk of spoken response, we use the remaining free time to generate the unspoken reasoning tokens. When a chunk of audio is played to the user, the model continues to generate the next unspoken reasoning chunk, achieving simultaneous thinking and talking. Remarkably, Stitch matches the latency of baselines that cannot generate unspoken CoT by design while outperforming those baselines by 15% on math reasoning datasets; Stitch also performs equally well on non-reasoning datasets as those baseline models. Some animations and demonstrations are on the project page: https://d223302.github.io/STITCH.

Real-time Automatic Speech Recognition (ASR) is a fundamental building block for many commercial applications of ML, including live captioning, dictation, meeting transcriptions, and medical scribes. Accuracy and latency are the most important factors when companies select a system to deploy. We present WhisperKit, an optimized on-device inference system for real-time ASR that significantly outperforms leading cloud-based systems. We benchmark against server-side systems that deploy a diverse set of models, including a frontier model (OpenAI gpt-4o-transcribe), a proprietary model (Deepgram nova-3), and an open-source model (Fireworks large-v3-turbo).Our results show that WhisperKit matches the lowest latency at 0.46s while achieving the highest accuracy 2.2% WER. The optimizations behind the WhisperKit system are described in detail in this paper.

Marking the end of a longtime tradition, the Fourth of July celebration at the Rose Bowl in Pasadena will not feature a fireworks show this year. Instead, there will be a drone show. The move comes as some venues have switched from fireworks to drone shows -- in which a fleet of drones performs a choreographed light show -- to celebrate the 4th of July. But drone shows have fallen flat for some. Notably Redondo Beach and Laguna Beach switched back to fireworks after trying out drone shows, and some promoters of fireworks shows have voiced criticism over efforts to transition to drone shows.

Large Language Models (LLMs) have demonstrated remarkable general capabilities, but enhancing skills such as reasoning often demands substantial computational resources and may compromise their generalization. While Parameter-Efficient Fine-Tuning (PEFT) methods offer a more resource-conscious alternative, they typically requires retraining for each LLM backbone due to architectural dependencies. To address these challenges, here we propose Universal Reasoner (UniR) - a single, lightweight, composable, and plug-and-play reasoning module that can be used with any frozen LLM to endow it with specialized reasoning capabilities. Specifically, UniR decomposes the reward into a standalone reasoning module that is trained independently using predefined rewards, effectively translating trajectory-level signals into token-level guidance. Once trained, UniR can be combined with any frozen LLM at inference time by simply adding its output logits to those of the LLM backbone. This additive structure naturally enables modular composition: multiple UniR modules trained for different tasks can be jointly applied by summing their logits, enabling complex reasoning via composition. Experimental results on mathematical reasoning and machine translation tasks show that UniR significantly outperforms existing baseline fine-tuning methods using the Llama3.2 model. Furthermore, UniR demonstrates strong weak-to-strong generalization: reasoning modules trained on smaller models effectively guide much larger LLMs. This makes UniR a cost-efficient, adaptable, and robust solution for enhancing reasoning in LLMs without compromising their core capabilities. Code is open-sourced at https://github.com/hangeol/UniR

Users often interact with large language models through black-box inference APIs, both for closed- and open-weight models (e.g., Llama models are popularly accessed via Amazon Bedrock and Azure AI Studio). In order to cut costs or add functionality, API providers may quantize, watermark, or finetune the underlying model, changing the output distribution -- often without notifying users. We formalize detecting such distortions as Model Equality Testing, a two-sample testing problem, where the user collects samples from the API and a reference distribution and conducts a statistical test to see if the two distributions are the same. We find that tests based on the Maximum Mean Discrepancy between distributions are powerful for this task: a test built on a simple string kernel achieves a median of 77.4% power against a range of distortions, using an average of just 10 samples per prompt. We then apply this test to commercial inference APIs for four Llama models, finding that 11 out of 31 endpoints serve different distributions than reference weights released by Meta.