What makes AI science explanations work? Start with bicycle gears

A gear change can make a climb feel easier. But should the larger sprocket be at the front or the rear? And does needing less force also mean using less energy? These questions give a spinning animation something concrete to explain.

V2EX author int64ago described building vistep with Astra. Following that lead, we explored its bicycle-gearing work. The current page offers a chaptered demonstration and controls for tooth counts, cadence and gradient. We will use it for a few comparisons, alongside a smaller experiment that isolates the drive ratio.

The author’s account of tool use does not establish that every part of the current page was produced independently by the model. We examine how one work explains a mechanism, without measuring learning outcomes or rating Astra’s overall teaching ability.

• Open the bicycle experiment ↗vistep · interactive workAfter loading, choose Explore to adjust gearing, cadence and gradient.
• The author’s build post ↗V2EX · 2026-09-06The account of using Astra, without attributing every implementation detail to it.

Imagine 34 positions around the front chainring. Advancing one tooth moves the chain by one pitch, so a full front revolution advances 34 pitches. A 12-tooth rear sprocket turns 34÷12 times to receive that chain; a 24-tooth sprocket turns 34÷24 times. This assumes no slip and a rear wheel engaged with the drive.

With the front chainring fixed, a larger rear sprocket gives fewer wheel turns per crank turn. At the same cadence, the bicycle travels more slowly. To explain what a larger sprocket does, first specify which end it is on and what stays constant.

• Exploratorium · Drives & Gears ↗Science-museum teaching materialExplains the rotation ratio by counting teeth.

Torque describes a force’s turning effect. Hold the wheel’s torque demand fixed and ignore drive losses: the mechanical work for one wheel revolution is unchanged. Distribute that work over more crank turns and less crank torque is needed. The small experiment changes turns and torque together, without creating energy.

Changing from 34/12 to 34/24 doubles the rear tooth count. At the same wheel load, required crank torque halves, while the crank must turn twice as often for the same wheel rotation. With the same crank length and force direction, lower torque corresponds to lower tangential pedal force.

Reaching the same height still increases gravitational potential energy according to mass and height. Real cycling also involves speed, air resistance, losses and human efficiency. A low gear can ease each pedal stroke without implying an equal proportional reduction in total energy use.

• OpenStax · Work and Power for Rotational Motion ↗Physics textbookRotational work and power; our small experiment applies idealized conditions.

Open vistep’s Explore mode and reset. Keep the 34-tooth chainring, 60 rpm cadence and 4% gradient, changing only the rear sprocket. On September 7, desktop Chrome displayed the three results below. They are rounded outputs of the webpage’s model, not measurements from a bicycle sensor.

From 12 to 24 teeth, speed approximately halves, but pedal force does not halve exactly. That need not indicate an incorrect ratio: the webpage also computes air resistance from speed, so lower speed reduces the load. It combines gearing with road load, while our earlier experiment holds the load fixed to answer a simpler question.

• Inspect the work’s calculation ↗Author’s source · pinned revisionSpeed feeds into air resistance, then wheel torque, crank torque and pedal force.

Try two more observations. Hold 34/32 gearing and 60 rpm, raising the gradient from 4% to 8%. In our observation, speed stayed at 8.2 km/h while pedal force rose from 83 N to 152 N. The model asks how much force is required to maintain the specified speed; a steeper road does not automatically slow its readout. A real rider may instead be unable to maintain cadence.

Then keep the gearing and 8% gradient, increasing cadence from 60 to 100 rpm. We observed an unchanged ratio of 1.06, speed rising to 13.6 km/h and required power increasing from 162 W to 279 W. Gearing sets the ratio of rotations; cadence sets their pace. These are different controls.

This suggests a reusable method for making explanations with AI: state the causal relationship, let readers change one input at a time, and make the image, numbers and explanation respond to the same change. When a prediction differs, explain the missing condition. Each control should have a clear question to answer.

Afterward, ask yourself why a larger front chainring differs from a larger rear sprocket, and why the model can keep its speed when the gradient increases. Explaining those points uses the relationships behind the picture, beyond memorizing a number.

• Continue exploring the original ↗vistep · browser experimentWe checked selected desktop interactions and pinned formulas, without evaluating the whole site, narration or learning outcomes.