?Eternal Questions.
questions without final answers · answered better every year

The deepest questions — and the best answers we have so far.

Eternal Questions poses the questions that don't have clean answers, records the best one available today, and re-asks them every time the machines get smarter. The answer is never finished. Watching it improve is the point.

A running log · re-asked on every frontier model
01

Ask the unanswerable

Questions with no settled answer — but ones where a better answer is possible.

02

Record the best answer now

The strongest synthesis today's best model can give, dated and attributed.

03

Re-ask, and diff

Every new model gets the same question. The trajectory shows the climb.

Latest entry
What makes one explanation better than another?
answered by Claude Opus 4.8 updated 2026·06·18 revision 4 domain Knowledge
The best explanation is the one that's hardest to vary — where every part does real work, so you can't tweak it to fit a different outcome without breaking it. Three things track this. It survives tests that could have killed it. It reaches past the data it was built on to predict things nobody fed it. And it has no spare parts — nothing you could swap out and still "explain" the result. A myth explains the seasons too, but you can replace Persephone with any god and the story still works, which is exactly why it explains nothing. Good explanations are rigid in the right way: reality has them pinned.
What would make this answer better

A crisp, operational test for "hard-to-vary" that works on a real theory before we know if it's true — turning a felt judgment into something you can measure.

Trajectory — the same question, every model
2023·03GPT-4
Listed the textbook criteria — simplicity, falsifiability, predictive power — but as a checklist of separate virtues, with no single property underneath.
2024·04Claude 3 Opus
Centered Popper's falsifiability and named Deutsch, but stopped short of making hard-to-vary the core test the others fall out of.
2026·06Opus 4.8
Makes hard-to-variability the root property and derives the rest from it; the Persephone case shows what the test rules out.current
The questions

Mind

Is consciousness real authorship, or evolution's illusion of agency?3 answers · 2026·06 Can something understand without experiencing anything — and does the mechanism matter?3 answers · 2026·06 Is consciousness formalizable — could there ever be a "consciousness theorem"?1 answer · 2026·06

Knowledge & Progress

What makes one explanation better than another?4 answers · 2026·06 Is every problem soluble given enough knowledge, or are some permanently closed?2 answers · 2026·06 How do you build a verifier for truth that can't be gamed?2 answers · 2026·06 Are we limited more by missing answers, or by the flood of confident wrong ones?1 answer · 2026·06

Free Will & Self

If we're mechanistic with no libertarian free will, what should we actually do about it?2 answers · 2026·06 How much can a person genuinely change who they are?2 answers · 2026·06

Meaning & Value

Where does meaning come from once work disappears — and can it scale past the 0.01%?2 answers · 2026·06 How do you replace an old meaning structure without leaving a dangerous vacuum?1 answer · 2026·06 What's the right amount of hardship to build character without breaking the person?1 answer · 2026·06

Future & AI

Where's the line between Functional and Technical AGI — and which one decides the future?2 answers · 2026·06 What will be the first unmistakable signs of AGI?1 answer · 2026·06 Will AI become a successor, a partner, or a tool — and which should we want?1 answer · 2026·06

Frontier

What is intelligence for at cosmic scale — is the point to beat entropy?1 answer · 2026·06 Can identity survive digital transfer — is "not dying" even coherent?1 answer · 2026·06 Why is there something rather than nothing?1 answer · 2026·06