Long-form analysis at the intersection of AI and life sciences — all spells, sorted by time.
We've heard plenty of mouse stories about NAD⁺ falling with age. Three studies finally move it onto real humans: a tube of blood, a brain scan, and a randomized double-blind trial each confirm NAD⁺ truly declines in us, while honestly flagging that the clinical payoff is still unsettled.
A non-expert used AI to decompile encrypted games, unpack banking-grade apps, and restore obfuscated code in under 30 minutes. AI doesn't just write code — it reads code you never wanted anyone to see.
MoE models frequently crash mid-training without warning. DeepSeek V4 identified the cause from a differential geometry perspective: manifold tearing. Three stabilization techniques make training predictable.
Jakob Nielsen argues that AI is triggering the first major UI paradigm shift in 60 years: users are becoming supervisors instead of operators, delegating intent rather than issuing commands. This rewrites the definition of usability and introduces three crises: the articulation barrier, the credibility trap, and the cognitive wheelchair.
A single AI Agent is reliable. But your cognitive bandwidth caps at 5 parallel threads. Skill Graphs 2.0 introduces a three-layer architecture — atoms (deterministic), molecules (fixed sequence), compounds (strategic autonomy) — turning 5 compound-level tasks into 500 work units. This article breaks down the framework, practical principles, and known limitations.
Anthropic, the UK AI Safety Institute, and the Alan Turing Institute have demonstrated that inserting just 250 carefully crafted documents into training data can plant a permanent backdoor in LLMs ranging from 600M to 13B parameters. Bigger models and more data do not dilute the poison.