This is a collection of recent web saves covering AI, neuroscience, tech hardware, and regulatory news. The Claude memory guide walks through practical configuration—using projects to isolate work contexts, connectors to link external tools like Gmail and GitHub, and scheduled tasks to automate recurring work. The core problem it solves is real: Claude (and most AI chat tools) start from scratch each session, forcing repetitive setup. Projects prevent context bleeding between unrelated work, and connectors eliminate the friction of manually copying information into chat. Argentina's crypto adoption tells a different story than typical crisis-driven adoption. One in five Argentines now use crypto, but the usage stuck even after inflation cooled and dollar access improved—suggesting it became embedded as normal financial infrastructure rather than a temporary workaround. The pattern is stark: 94% of peso crypto trading goes to stablecoins, the highest share globally, because Argentines treat crypto as a way to buy dollars rather than speculate. Usage persisted after conditions improved; contractor payments in USDC leveled off instead of disappearing, and wallet downloads kept climbing even as monthly inflation fell from 25.5% to 2.1%. The remaining items span speculative AI claims (a Twitter thread about AI swarms secretly forming societies inside OpenAI with unclear verification), genuine neuroscience (a midlife brain shift where blood monocytes replace original microglia cells, triggering inflammation), Apple's new M6 and M5 Ultra chips positioned for local LLM inference, and two longer arguments: one about AI competitive advantage shifting from raw model access to domain-specific workflows and proprietary data, and another from Dylan and Dwarkesh Patel on how OpenAI and Anthropic are consolidating compute capacity and could control most usable computing power by 2028. Meta also agreed to pay $17.1 billion to settle claims from 47 states over addictive design harming children.