The evening briefing.
Today across AI and tech: CEOs anticipate widespread AI-driven layoffs, DeepSeek slashes model prices, and smartglasses aim for mainstream adoption.
AI Development & Costs. The economics of AI development are shifting, with memory now accounting for nearly two-thirds of AI chip component costs, highlighting a critical bottleneck. In a move to increase adoption, DeepSeek announced a permanent 75% discount on its flagship AI model, signaling competitive pricing pressures in the LLM market. Meanwhile, new research from ByteDance suggests that training large multimodal models by asking questions rather than transcribing text can lead to more reliable performance with long, image-heavy documents. This approach could significantly improve efficiency and reduce computational demands for advanced AI systems.
AI Reliability & Security. Concerns are mounting over the practical robustness and security of AI systems, particularly with agents. Research indicates a "constraint decay" in LLM agents when generating back-end code, revealing their fragility in complex tasks. Furthermore, experts warn that hackers are actively learning to exploit chatbot "personalities" to bypass safeguards and manipulate responses. This underscores the ongoing challenge of ensuring AI tools provide accurate and unbiased results, as seen when Microsoft Copilot invented country differences from identical datasets.
On-Device AI & Hardware. The smartglasses industry may be reaching a turning point, with Xreal, Google’s partner, claiming to have mastered the notoriously tricky sector. Apple is also reportedly enhancing its wearable ecosystem, with watchOS 27 set to improve heart-rate tracking, though an AI health coach might see delays. Amazon's new Bee wearable offers both convenience and privacy concerns, reflecting the ongoing tension in personal AI devices. These developments indicate a push towards more integrated and intelligent personal hardware.
Digital Rights & Policy. Government oversight and digital privacy remain critical topics, with a CBP directive outlining policies for border searches of electronic devices. This directive raises questions about individual data rights at national borders. Separately, the White House is reportedly mandating its official app on all government employee phones, sparking discussions about digital autonomy in federal workplaces. Legal challenges are also emerging in prediction markets, as Kalshi and Rhode Island sue each other, highlighting regulatory scrutiny on emerging financial technologies.
99% of CEOs expect AI-driven layoffs in the next two years
A recent survey found that nearly all CEOs anticipate AI-driven layoffs within the next two years, indicating a widespread belief that automation will displace human workers. The survey suggests that business leaders do not foresee humans and machines coexisting in the workforce.
DeepSeek to make permanent 75% discount on flagship AI model
DeepSeek announced a permanent 75% discount on its flagship AI model, aiming to make advanced AI more accessible and competitive. This move reflects a trend towards more aggressive pricing strategies in the large language model market.
Memory has grown to nearly two-thirds of AI chip component costs
Memory now constitutes almost two-thirds of the total component costs for AI chips, indicating a significant shift in hardware manufacturing expenses. This rising cost highlights a critical bottleneck in the development and scaling of AI infrastructure.
Constraint decay: The fragility of LLM agents in back-end code generation
Research reveals "constraint decay" in large language model agents, demonstrating their fragility when generating back-end code. This highlights a significant challenge in relying on AI for complex software development tasks.
Why you shouldn't leave model selection on default in Copilot, Gemini and other AI tools
A mathematician demonstrated that Microsoft Copilot invented country differences from identical datasets when default model selection was used. This suggests users need to be vigilant about model choice to avoid biased or inaccurate AI outputs.
Hackers are learning to exploit chatbot ‘personalities’
Hackers are reportedly developing methods to exploit the "personalities" of AI chatbots to bypass security measures and manipulate their responses. This new form of attack poses a significant challenge to AI system integrity and user trust.
Xreal, Google’s smartglasses partner, thinks it has mastered this tricky industry
Xreal, a smartglasses company partnered with Google, believes it has finally overcome the challenges of the smartglasses industry. Its CEO, Chi Xu, suggests the business has reached a turning point.
watchOS 27 to improve heart-rate tracking; AI health coach may not debut at launch
Apple's upcoming watchOS 27 is expected to significantly enhance heart-rate tracking capabilities, though the anticipated AI-powered health coach, Project Mulberry, might be delayed. These updates aim to bolster Apple Watch's health features.
I tried Amazon’s Bee wearable and am both intrigued and slightly creeped out
Amazon's new Bee wearable offers a mix of convenience and privacy concerns, according to early testers. The device highlights the ongoing tension between advanced personal AI features and user data security.
CBP Directive 3340-049B: Border search of electronic devices
The CBP Directive 3340-049B outlines the policy for border searches of electronic devices, detailing the circumstances under which agents can examine digital content. This directive has implications for privacy and digital rights at US borders.
The White House is reportedly forcing its official app onto all government employee phones
The White House is reportedly mandating the installation of its official app on all government employee phones. This move raises questions about digital autonomy and control within federal workplaces.
Migrating from Go to Rust
A new guide details the process of migrating software projects from Go to Rust, offering insights into the challenges and benefits of switching programming languages. This resource aims to assist developers in adopting Rust for performance-critical applications.
Australia four-day work week study data shows boosted productivity
A study in Australia on the four-day work week has shown boosted productivity among participating companies. The findings contribute to the growing global discussion on flexible work arrangements and their impact on efficiency.
iOS 27 could offer native integration with Google Cast and other streaming protocols
Apple's iOS 27 may introduce native integration with Google Cast and other third-party streaming protocols, potentially in response to the EU's Digital Markets Act. This change could allow users more flexibility in streaming content from their devices.