Scaling LLM vocabulary is often used to reduce input sequence length and alleviate attention's quadratic cost. Yet, current LLM architectures impose a critical bottleneck to this procedure: the output projection layer scales linearly with vocabulary size, rendering substantial expansion impractical. We propose SpeLLM, a method that decouples input and output vocabularies by predicting character-level strings through multiple output heads. In SpeLLM, each of the $k$ linear heads predicts a single character simultaneously, enabling the model to represent a much larger output space using smaller, independent linear heads. We present a self-distillation approach for converting a standard LLM to a SpeLLM. Our experiments with four pre-trained LLMs show their SpeLLM variants achieve competitive performance on downstream tasks while reducing runtime by 5.1% on average across models. Our approach provides a potential avenue for reducing LLM costs, while increasing support for underrepresented languages and domains.

A young entrepreneur who was among the earliest known recruiters for Elon Musk's so-called Department of Government Efficiency (DOGE) has a new, related gig--and he's hiring. Anthony Jancso, cofounder of AcclerateX, a government tech startup, is looking for technologists to work on a project that aims to have artificial intelligence perform tasks that are currently the responsibility of tens of thousands of federal workers. Jancso, a former Palantir employee, wrote in a Slack with about 2000 Palantir alumni in it that he's hiring for a "DOGE orthogonal project to design benchmarks and deploy AI agents across live workflows in federal agencies," according to an April 21 post reviewed by WIRED. Agents are programs that can perform work autonomously. "We've identified over 300 roles with almost full-process standardization, freeing up at least 70k FTEs for higher-impact work over the next year," he continued, essentially claiming that tens of thousands of federal employees could see many aspects of their job automated and replaced by these AI agents.

The'In Case of Emergency' shortcut can send a message to your emergency contacts. For some reason, Siri responds about 60% of the time when I ask it to call or text someone. Yes, this garbage even happens to me, and I've been helping folks with their tech lives for decades. That's why I've channeled my frustrations into easy ways to make your tech less annoying, too. No purchase necessary to enter.

It has been lightly edited. When a young Chinese man sat down at his QWERTY keyboard in 2013 and rattled off an enigmatic string of letters and numbers, his forty-four keystrokes marked the first steps in a process known as "input" or shuru. Shuru is the act of getting Chinese characters to appear on a computer monitor or other digital device using a QWERTY keyboard or trackpad. The young man, Huang Zhenyu, was one of around 60 contestants in the 2013 National Chinese Characters Typing Competition. But Zhenyu's prizewinning performance wasn't solely noteworthy for his impressive typing speed--one of the fastest ever recorded.

Computer science has a language problem--and we are not alluding to programming languages. Many prevalent, flawed views about natural human language are limiting who is in computer science and what people can accomplish with the technology we build. To start, computer science centers around the English language, and that produces technologies that work poorly for many people. As Manuel Pérez-Quiñonesa points out, when developers make assumptions about English as the default language, navigating digital device interfaces can be frustrating, even for a professional computer scientist fluent in English such as Pérez-Quiñones. Poor multilingual or character-encoding support, incorrect cultural norms baked into software, and so on--these challenges confront users all over the world.

Google has begun testing new features such as autocorrect, improved file search, and even adaptive charging within the latest version of Chrome OS, meaning those features will arrive on the stable versions of Chromebooks soon. The new features are part of Chrome OS M116, available now for download and testing. Autocorrect: Autocorrect is now enabled by default for those apps that use English as a primary language, automatically fixing typos, spelling, and other errors. Grammar, however, is not affected. Improved search within the Files app: You can now search across both your local files as well as Google Drive at the same time.

Windows 11 has a lot of features you may not know about. CyberGuy shows you how to customize your computer. We recently got a question from Wayne from Burgettstown, Pennsylvania. "I would like tips on using Windows 11. CLICK TO GET KURT'S FREE CYBERGUY NEWSLETTER WITH SECURITY ALERTS, QUICK TIPS, TECH REVIEWS AND EASY HOW-TO'S TO MAKE YOU SMARTER It's been about 20 months since Windows 11 was released, and its capabilities are pretty impressive.

After more than a decade, autocorrect "fails" could be on their way out. Apple's much-maligned spelling software is getting upgraded by artificial intelligence: Using sophisticated language models, the new autocorrect won't just check words against a dictionary, but will be able to consider the context of the word in a sentence. In theory, it won't suggest consolation when you mean consolidation, because it'll know that those words aren't interchangeable. The next generation of autocorrect was one of several small updates to the iPhone experience that Apple announced earlier this month. The Photos app will be able to differentiate between your dog and other dogs, automatically recognizing your pup the same way it recognizes people who frequently appear in your pictures.

Is that what autocorrect is doing? It all seems so mysterious. What's going on inside that artificial construct? Maybe there are acres of intense cores all zapping away, spitting billions of tiny sparks of lightning through the intricacies of its complex interior. One of my theories is that it's learning by annoying harmless people like me.

Even as virtually every aspect of smartphones has gotten at least incrementally better over the years, autocorrect seems stuck. An iPhone 6 released nearly a decade ago lacks features such as Face ID and Portrait Mode, but its basic virtual keyboard is not clearly different from the one you use today. This doesn't seem to be an Apple-specific problem, either: Third-party keyboards can be installed on both iOS and Android that claim to be better at autocorrect. Disabling the function altogether is possible, though it rarely makes for a better experience. Autocorrect's lingering woes are especially strange now that we have chatbots that are eerily good at predicting what we want or need.