Community
Knowledge worth keeping, shared
People share whole volumes from their own Eigma here. Look one over, and when it fits, bring it into yours.
Humanities
Why AI's Builders Asked to Be Restrained
In September 2026, Anthropic CEO Dario Amodei published an essay asking the AI industry to slow down, and Sam Altman and Elon Musk agreed within days. This volume treats that moment as a doorway: the control problem inside it, why no lab can pace itself alone, why rivals can't verify restraint, and four questions that outlast the news cycle.
Contents 5 chapters
- The Paradox of the Pause 2p
- The Two Curves of Development 2p
- Saying What You Want 2p
+ 2 more
Law
The EU AI Act in Everyday Life
Since August 2026 a system you talk to has to tell you it is a machine, and a few practices are banned outright. Almost everything else waits: the safeguards for AI in hiring, credit, benefits and borders are scheduled for December 2027, and some dates have already moved once. Meanwhile the law that actually helps you today is often not this one.
Contents 5 chapters
- Encounters with AI 4p
- Work and education 4p
- Money and essential services 4p
+ 2 more
AI & ML
How AI Changed Fraud
No complaint system records whether AI was involved, so anyone quoting you a share of fraud losses invented it. What can be said: a familiar voice is now cheap to manufacture, a live video call establishes less than it appears to, and the defences that hold sit on channels, authentication and payments rather than on spotting a fake.
Contents 5 chapters
- The measured problem 4p
- Synthetic methods 4p
- Victims and circumstances 4p
+ 2 more
Science
Forgetting and Spaced Repetition
Spacing beats cramming, and retrieving an answer teaches more than reading it again. Those two findings are among the most replicated in psychology. Almost everything built on top of them is not: no validated interval, no daily card target, no evidence that a newer algorithm remembers you better. This volume draws that line and stays on the right side of it.
Contents 5 chapters
- How forgetting works 4p
- What actually strengthens memory 4p
- Scheduling 4p
+ 2 more
AI & ML
What Is Inside a Chatbot
Your sentence is cut into tokens, and nearness between their positions is what the machine has instead of meaning. From there it predicts what comes next, which is why it has nothing to cite: the answer is composed, not retrieved. It sounds equally certain when it is wrong, and the reason it cannot count letters is not the one usually given.
Contents 5 chapters
- What the thing is 4p
- Where the ability comes from 4p
- Why it gets things wrong 4p
+ 2 more
AI & ML
Coding Agents From First Principles
The context window is a budget you spend, not a memory that remembers, and most of what fills it was never chosen for the task. A project brief changes what an agent does but cannot compel it. A task is finished when something machine-checkable says so, not when the agent says so. The ways this goes wrong have names, symptoms and recoveries.
Contents 5 chapters
- Why agents fail 4p
- Workspace setup 4p
- Working a task 4p
+ 2 more