Learning routes

Fourteen routes that end in something you make

A route is a question with a fixed number of stops, drawn here to the time each stop takes. The square at the end is the thing you keep. Each route is also a file you can download and change.

• Can you reproduce a published AI safety result yourself?

  7 stops · 2h30m alone, 45m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a tested practice

  A one-page reproduction report of a published AI safety result, with the figures you obtained, a second person's rerun, and the standard it was checked against.

• How fast is the computing power used to train frontier AI growing?

  7 stops · 3h20m alone, 45m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a tested practice

  A one-page note reproducing Epoch AI's growth rate for frontier training compute from its own data, with a second person's recomputation, the counterpoint, and the standard it was checked against.

• Would you recognise an instruction hidden in text that an AI reads?

  7 stops · 3h alone, 45m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a tested practice

  A labelled test set of nine harmless prompt-injection texts, one for each scenario in OWASP's entry, with the safe behaviour for each, a second person's labels, and the standard it was checked against.

• Can a simple law predict when a network suddenly starts to generalise?

  7 stops · 3h20m alone, 45m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a tested practice

  A one-page note reproducing the exponents of a published grokking delay law from the release's own table, with a second person's recomputation, the claim's limits, and the standard it was checked against.

• Can you write an AI evaluation that someone else can rerun?

  7 stops · 4h alone, 45m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a tested practice

  A small evaluation in Inspect with its code, requirements and README, scored against BetterBench's criteria, and a record of whether a second person could rerun it and got the same scores.

• What does an AI company's safety policy actually commit it to?

  8 stops · 4h25m alone, 45m together

  Can correct you: stop 5, an outside counterpointAway from the screen: stop 3, a tested practice

  A one-page reading of one AI company's safety policy: what it commits to, whether each commitment could be checked, and a grade on each of a published rubric's seven criteria, with a second person's grades beside yours.

• What could brain science offer to making AI safe?

  6 stops · 3h35m alone, 45m together

  Can correct you: stop 3, an outside counterpointAway from the screen: stop 4, a making

  A one-page note turning one neuroscience path to AI safety into a testable claim, with the experiment, the result that would count against it, a published counterpoint and a scientist's objection.

• How much will AI add to economic growth, and why do the estimates differ so much?

  7 stops · 3h45m alone, 45m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a tested practice

  A one-page table setting two published estimates of AI's effect on productivity side by side, input by input, with your own recomputation, the input that moves the answer most and an economist's objection.

• How well have forecasters predicted AI, and how would you score your own forecasts?

  7 stops · 3h55m alone, 45m together

  Can correct you: stop 3, an outside counterpointAway from the screen: stop 4, a tested practice

  A forecast sheet of ten questions about AI with your probabilities, reasons and resolution dates, a second person's forecasts beside them, and a column that scores both when the answers arrive.

• Do AI models make biological misuse easier, and how would anyone know?

  7 stops · 3h50m alone, 45m together

  Can correct you: stop 3, an outside counterpointAway from the screen: stop 4, a tested practice

  A one-page reading note setting two studies of AI and biological misuse against a published reporting standard, with the change that would most improve each design and a reviewer's objections.

• Where should someone new to AI safety start?

  6 stops · 5h15m alone, 1h30m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 5, a person

  A one-page note: what you now think about AI risk, the strongest point against it, and one thing you will do next.

• Could local AI strengthen community without increasing surveillance?

  6 stops · 3h alone, 90m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 2, a place

  A two-page design: the need, the smallest useful function, what it knows, what it refuses to know, who is responsible for what, how it fails, and how a person leaves it.

• What makes a recurring gathering outlive its founder?

  6 stops · 2h30m alone, 60m together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a person

  A one-page continuity plan the gathering can adopt, amend, or refuse.

• What does one intentional community near you actually invite?

  6 stops · 2h alone, half a day, if a visit is invited together

  Can correct you: stop 4, an outside counterpointAway from the screen: stop 3, a place

  A dated verification note in the map's entry format, sent through the map's own correction form. Nobody but its maker has checked an entry on the map yet.

What a route is

1. 1

   Your question

   You start from what you want to know, and say what you have to work with: time, energy and whether you are learning alone or with others. Route notes stay in this browser until you clear its site data.

   Built

2. 2

   A few steps, ranked in the open

   Each step offers options in an order you can check: the standards behind it, where each option comes from, and weights you can change.

   Shown on one worked example below

3. 3

   Something that can correct you

   At least one step can prove you wrong: an independent source, a practitioner, or a test. A challenge has to come from outside whatever wrote the route.

   Built

4. 4

   Somewhere away from the screen

   At least one step happens with a person, in a place, or with your hands.

   Built

5. 5

   Work you keep

   Every route ends in something you made, kept with its sources: a design, a plan, a verified correction, a changed question.

   Built

6. 6

   Work that counts elsewhere

   Checked against a published standard, so a school, workshop or employer could recognise it. Some routes name their standard and let you check your figures; no partner recognises the work yet.

   Shown on 9 routes

From work you keep to evidence someone else can assess

One step, ranked in the open

No route ranks its options yet, so here is one step worked through: seven real things to look at for a first picture of synchrony, in the order the standards put them. Change what you have, or how much each standard counts, and the order changes with it.

Also on Kosplora

• Scientific simulations: twenty ideas from mathematics and science, each running in your browser with four short notes and its sources.
• Patterns across subjects: 17 rules that turn up in places that look nothing alike, starting with synchrony.
• Educational resource library: courses, reference works and guides from Values Commons, by subject.