Transformer-based Large Language Models (LLMs) have demonstrated powerful in-context learning capabilities. However, their predictions can be disrupted by factually correct context, a phenomenon known as context hijacking, revealing a significant robustness issue. To understand this phenomenon theoretically, we explore an in-context linear classification problem based on recent advances in linear transformers. In our setup, context tokens are designed as factually correct query-answer pairs, where the queries are similar to the final query but have opposite labels. Then, we develop a general theoretical analysis on the robustness of the linear transformers, which is formulated as a function of the model depth, training context lengths, and number of hijacking context tokens. A key finding is that a well-trained deeper transformer can achieve higher robustness, which aligns with empirical observations. We show that this improvement arises because deeper layers enable more fine-grained optimization steps, effectively mitigating interference from context hijacking. This is also well supported by our numerical and real-world experiments. Our findings provide theoretical insights into the benefits of deeper architectures and contribute to enhancing the understanding of transformer architectures.

Proprietary large language models (LLMs) exhibit strong generalization capabilities across diverse tasks and are increasingly deployed on edge devices for efficiency and privacy reasons. However, deploying proprietary LLMs at the edge without adequate protection introduces critical security threats. Attackers can extract model weights and architectures, enabling unauthorized copying and misuse. Even when protective measures prevent full extraction of model weights, attackers may still perform advanced attacks, such as fine-tuning, to further exploit the model. Existing defenses against these threats typically incur significant computational and communication overhead, making them impractical for edge deployment. To safeguard the edge-deployed LLMs, we introduce CoreGuard, a computation-and communication-efficient protection method. CoreGuard employs an efficient protection protocol to reduce computational overhead and minimize communication overhead via a propagation protocol. Extensive experiments show that CoreGuard achieves upper-bound security protection with negligible overhead.

The Download: "reprogramming" aging, and the hidden sense of interoception Plus: SpaceX has officially delivered the largest IPO in history. Why "reprogramming" is the buzziest approach to reversing aging right now Earlier this week, Life Biosciences, a biotech company focused on reversing age-related diseases, announced that it had dosed its first volunteer. A person with glaucoma has had an experimental treatment injected straight into their eyeball. The idea is to treat the disease by regenerating healthy nerves in the eye--but the company already hopes to go further. If the treatment can reverse glaucoma, similar treatments could reverse other diseases of aging. Maybe, just maybe, they could reverse aging altogether.

When you purchase through links in our articles, we may earn a small commission. The Nintendo Switch 2 almost never goes on sale, but a new Woot coupon finally knocks $15 off the price. The Nintendo Switch 2 is rarely discounted, but you can save $15 on your purchase right now. That's not a huge discount, sure, but on a console that never goes on sale, it's a fantastic excuse to finally take the plunge. The Nintendo Switch 2 launched in June 2025, and since then, we haven't seen it on sale despite checking frequently.

Large language models (LLMs) have revolutionized natural language processing and are increasingly applied to other sequential data types, including genetic sequences. However, adapting LLMs to genetics presents significant challenges. Capturing complex genomic interactions requires modeling long-range global dependencies within DNA sequences, where interactions often span over 10,000 base pairs, even within a single gene. This poses substantial computational demands under conventional model architectures and training paradigms. Additionally, traditional LLM training approaches are suboptimal for DNA sequences: autoregressive training, while efficient for training, only supports unidirectional sequence understanding. However, DNA is inherently bidirectional.

The first ever halftime show at a FIFA World Cup final has been curated by Coldplay's Chris Martin and will feature performances from global superstars.

Nature is always performing chemistry experiments, and in the dark and sticky corners of its forests and jungles, it creates compounds that have hyper-specific effects on the human mind. Many people of different ages and cultural backgrounds have eaten this mushroom and experienced the same hallucination. They report seeing elf-like figures that parkour around on clothes, on furniture, and on walls. These little people seem to like dancing and performing acrobatics. Large groups of them will march in formation. This "lilliputian hallucination" can last for a day, and closing your eyes is no escape.

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?

PCWorld analyzes how Anthropic's powerful Fable 5 AI model faces accessibility issues and data retention controversies, while Apple's revamped Siri offers practical integration. Apple's AI features include iCloud data analysis, email composition, and Private Cloud Compute for privacy, making AI tools accessible to millions of users. Despite Fable 5's advanced capabilities, Apple's approach appears more likely to deliver meaningful AI benefits for everyday tasks and decision-making. Anthropic's first Mythos-class Claude model, Fable 5, hit the world like an atom bomb this week, and that's barely an exaggeration. But Apple's rebooted Siri could be the AI moment that actually reaches everyone else. A modified version of Mythos, the benchmark-shattering Claude model that's scary-good at cybersecurity and worryingly knowledgeable about bioweapons, Fable 5 comes wrapped in so many safeguards that it reportedly refuses even the most basic chats about biology .

Claire chatted to Edward Mehr from Machina Labs about their RoboCraftsman that shapes complex metal parts for the aerospace, defence, and automotive industries. Edward Mehr is an entrepreneur and engineer specializing in advanced manufacturing, robotics, and artificial intelligence. As the Co-Founder and CEO of Machina Labs, he leads efforts to integrate AI-driven robotics into flexible, on-demand production systems. Under his leadership, Machina Labs is reshaping how industries such as aerospace, defence, and automotive approach metal forming and modern manufacturing. Before founding Machina Labs, Ed worked at leading technology companies, including Relativity Space, Averon, SpaceX, Google, and Microsoft.