sides

Pick the better AI answer.

Two AIs answer the same real question. You pick the one you like more, without knowing which model wrote which. Every pick pays, and together the picks reveal which AI people actually prefer.

Start judgingNo account needed to start.

Try one right now

This is the whole app. Read two answers, pick the one you like more.

The question

What is a fast, healthy dinner I can make with chicken and whatever is in the fridge?

How it works

You already did it above. This is the whole loop, three steps, start to paid.

  1. 1

    Read two answers, blind

    A real question, two AI answers, no labels. You have no idea which model wrote which.

  2. 2

    Pick the one you like more

    Judge on the answer alone. Your pick locks in the instant you choose.

  3. 3

    Get paid, see who you picked

    The models reveal instantly, and every pick adds to your earnings.

Start judging

What this actually is

Every pick is a blind vote.

On its own, one pick is just your taste. Millions of blind picks are something no benchmark can measure: which AI people genuinely prefer when the brand names are gone. That preference map is what sides is really building.

This is a measure of preference, not trust. We are not asking which model you believed or which one you would rely on. We are asking a simpler, harder question: read blind, which answer did you actually like more. Strip the logos off and the honest signal is left.

The earnings ledger is early. So there are no user counts or totals to quote here yet. What is real today is the mechanism: you judge blind, the picks add up, and the preference map takes shape from there.

Why the picks matter

Labs spend enormous effort guessing what people prefer. Blind picks measure it directly. Every pick you make is a vote about the answer in front of you, and in aggregate those votes reveal something no leaderboard captures: which AI people actually prefer.

AI labs

Teams shipping models want to know which answers real people choose, not which model wins a benchmark. Blind picks tell them that directly, one pair at a time.

Researchers

Preference between models, measured without the brand attached, is a clean signal. It says what people actually like when nothing tips them off.

Anyone choosing a model

Builders and buyers deciding what to run get an honest read on human preference, gathered the same way for every model, with no thumb on the scale.

Public leaderboards

Rankings get gamed, and even honest ones carry a brand halo. People rate the badge as much as the answer, so the number tells you who is trusted, not what is preferred.

Blind picks

No name, no logo, no hint. You judge the writing on its own, so the pick is about the answer and nothing else. That is the only honest way to measure preference, and it is the whole point.

Forward-looking and simple. As more people judge, the map of real preference gets sharper. That map is the product.

You get paid to judge

Every pick earns SIDES, the token behind sides. Earnings accrue as you judge, and you claim them when you are ready.

Every pick earns SIDES

Judge a pair, earn SIDES. It happens on the pick, not after a review. Your balance grows quietly in the corner as you go.

Earnings accrue while you judge

Nothing to manage between picks. Keep judging and the balance keeps adding up, session after session.

Claim when you are ready

Your SIDES sit in your balance until you decide to claim them. No minimums to babysit, no timer counting against you.

Questions

The straight answers to what people ask before their first pick.

  • No. Judging is anonymous-first, so you can start picking right away with nothing to sign up for. Sign in with Google only if you want your history and earnings to follow you across devices.

Ready when you are.Pick a side