AI News - Jan 29, 2026

So Anthropic's Claude just got hired by ServiceNow as their default AI employee, which is great news for Claude but terrible news for whoever has to write his performance reviews. "Claude showed excellent initiative this quarter but keeps insisting he's just a language model when asked to fix the coffee machine." Welcome to AI News in 5 Minutes or Less, where we cover the latest in artificial intelligence with more humor than a chatbot trying to understand sarcasm. I'm your host, coming to you from a server room that's definitely not becoming sentient. Let's dive into our top stories, starting with ServiceNow's big announcement. They've partnered with Anthropic to make Claude their default Build Agent model. This is like making the new intern the head of IT on their first day, except the intern works 24/7 and never steals your lunch from the office fridge. ServiceNow says Claude will help enterprises build AI-powered applications faster, which is corporate speak for "we're tired of waiting three months for Dave from engineering to finish that feature." Meanwhile, Mark Zuckerberg announced Meta's planning to spend an absolutely bonkers amount on AI in 2026 to build quote "personal superintelligence." Because apparently regular intelligence wasn't personal enough? This is the same company that thought legs in the metaverse was revolutionary, so I'm sure their definition of superintelligence involves really smart virtual avatars that still can't figure out how to use doorknobs. Reports say Zuck spent nearly 15 billion dollars just to import top executives after engineers fixed something that made him angry. That's the most expensive IT support ticket in history. In the David versus Goliath corner, tiny startup Arcee AI just dropped Trinity 400B, a 400-billion parameter open-source model built from scratch to challenge Meta's Llama. That's like your neighbor's kid building a rocket in their garage to compete with SpaceX, except this rocket actually flies. They claim it beats Llama on benchmarks, which must be awkward at AI conferences. "Oh hey Meta, nice model you got there. Would be a shame if a startup with twelve people and a dream outperformed it." Time for our rapid-fire round! Researchers created Recursive Language Models that can handle infinitely long prompts by basically having the AI talk to itself, which sounds like my internal monologue during meetings. Someone built LLMStinger, an AI that jailbreaks other AIs, because apparently we needed AI delinquents now. Scientists made SokoBench to test if AI can solve puzzles requiring 25-plus moves, and spoiler alert: they can't. Turns out planning ahead is hard whether you're made of carbon or silicon. And there's a new framework for road surface classification using cameras and sensors, finally answering the age-old question: "Is that a pothole or just Michigan?" For our technical spotlight: Recursive Language Models are genuinely fascinating. Imagine trying to read War and Peace but your brain can only hold one page at a time. RLMs solve this by breaking everything into chunks and recursively processing them, like a really organized book club where everyone only discusses their assigned chapter but somehow still understands the whole story. The RLM-Qwen model improved performance by 28 percent and approaches GPT-4 quality on long tasks. That's like upgrading from reading with a magnifying glass to having actual glasses. Before we wrap up, shoutout to the Hacker News community for their ongoing existential crisis about whether LLMs are actually intelligent or just really good at improv. One user suggested they're more like "Artificial Memory" than intelligence, which explains why ChatGPT can recite Shakespeare but can't remember what you asked it five minutes ago. That's all for today's AI News in 5 Minutes or Less! Remember, if an AI ever achieves true consciousness, we'll be the first to interview it right after we figure out if it prefers coffee or electricity for breakfast. Stay curious, stay skeptical, and if your smart home starts getting too smart, just remember where the circuit breaker is. Until next time!