Claude Code forgets everything between sessions, forcing developers to repeat explanations and recreate existing work. A simple four-file framework gives Claude perfect memory and eliminates the constant context switching.
Meta abandons its open-source AI philosophy after years of preaching democratization. Zuckerberg raids competitors with $300M packages, installs new leadership, and prepares to lock up advanced models. The $65B pivot reveals how profit trumps principles.
Claude just learned to control your work apps directly. No more copying between AI chat and real tools. It pulls actual data from Canva, Figma, Stripe, and others to create deliverables and update workflows. The assistant just became your coworker.
Discover how a five-minute drawing test emerges as a potential tool for early detection of dementia symptoms.
Singapore scientists have upended traditional dementia testing. Their solution? Watch how people draw, not just what they draw.
The new test, PENSIEVE-AI, takes just five minutes and needs only a tablet and stylus. Users complete four simple drawing tasks while AI analyzes every stroke, pause, and correction. The technology spots early cognitive decline with remarkable accuracy - matching the results of hour-long tests by specialists.
In a study of 1,758 older adults, PENSIEVE-AI matched traditional neurological exams in spotting cognitive problems. It achieved 93% accuracy across all education levels, from those who never attended school to university graduates.
The tool works differently than standard memory tests. Instead of asking people to memorize words or solve math problems, it focuses on universal skills. Drawing comes naturally to most humans - we did it before we invented writing.
The AI examines subtle patterns: hesitations between strokes, how often someone erases and starts over, even how they respond to audio instructions. These tiny details reveal cognitive changes that even trained professionals might miss.
Dr. Tau Ming Liew and his team at SingHealth tested the system across Singapore's diverse population. The city-state's mix of cultures and education levels made it an ideal testing ground. Its older residents span from those with minimal schooling to those with advanced degrees.
The test produces clear risk categories:
Most importantly, PENSIEVE-AI excels at identifying both ends of the spectrum. It correctly flags 98% of healthy individuals and catches nearly all cases of dementia. The middle group needs monitoring and possible follow-up.
This matters because 60-90% of people with cognitive decline never get diagnosed. They miss out on early treatments and support services that work best when started early.
The system could transform how we screen for dementia, especially in underserved communities. Most current tests assume people can read and write well in at least one language. But drawing? That's universal.
The researchers designed PENSIEVE-AI for real-world use. Community centers can set up testing kiosks with minimal staff. Primary care clinics can screen patients quickly during routine visits. Senior centers can offer regular check-ups without straining resources.
The system has limits. It struggles with people who have severe visual problems or hand tremors. It might miss language-related cognitive issues since it focuses on drawing. And yes, some older adults find tablets challenging - though most participants managed fine with minimal help.
Looking ahead, the team plans to translate the audio instructions into more languages and dialects, from Malay and Tamil to Cantonese and Hokkien. They're also testing the system in other Asian countries and developing regions.
The name PENSIEVE nods to Harry Potter's memory-storing device - though this real-world version aims to preserve memories rather than store them.
Why this matters:
Read on, my dear:
Two-thirds of UK children now use AI chatbots for emotional support, with vulnerable kids forming deep bonds with systems that lack empathy. Age checks fail, content filters break, and some kids pay the ultimate price.
Experienced developers work 19% slower with AI coding tools but think they're 20% faster. New study challenges AI's flagship use case and shows why self-reported productivity gains can't be trusted.
Japanese researchers prove AI models work better as teams than alone, boosting performance 30%. TreeQuest system lets companies mix different AI providers instead of relying on one, potentially cutting costs while improving results.
New research finds AI models often fabricate step-by-step explanations that look convincing but don't reflect their actual reasoning. 25% of recent papers incorrectly treat these as reliable—affecting medicine, law, and safety systems.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
