Bitter T2R-001

T2R-001 / The taste model for ideas

Ideas Will Never Taste The Same.

T2R-001 / The taste model for startup ideas

AI gives you sugar.We bring the flavour.

Bitter is an AI model for startup ideas. It finds ideas other AIs don't suggest, checks them against the real world and shows you the evidence.

See how it works

How it works

  1. 1

    It maps the sugar

    It finds the ideas AI keeps handing everyone, so you can avoid them.

  2. 2

    It leaves the map

    It looks for ideas far from that crowded middle.

  3. 3

    It tastes the world

    Already built? Free in ChatGPT? Is anyone asking? Most ideas don't survive. You see the evidence for the ones that do.

Dead
cheap.

A drop of Bitter costs far less than the year you'd lose on the wrong idea. We publish the exact cost per idea with the first public run.

Why the name?

Five tastes.
AI has four.

Bitter is generally thought to have evolved to stop animals eating poison. Humans carry about 25 receptors for it, called T2Rs. AI shipped with none, so it swallows everything. T2R-001 is the missing receptor.

AI called all of these great.

Chatbots are built to agree: one Stanford study found they affirm users about 49% more than humans do. Here's what a real check found.

  • Antivirus for AIAlready built: Norton Safe AI, AIR ($50M).
  • Oral examinerChatGPT already does it.
  • Download more RAMToo narrow to be a company.
  • Customs cursor buddyA copy of something that exists.
  • Lawsuit filterNot worth a year of your life.
  • Third WheelNobody needs it.

Real ideas AI gave us. These checks were done by people. T2R-001 will do them for you.

AI Can Build Anything. It Can't Tell You What's Worth Building.

Anyone can ask a chatbot to critique an idea. We cause the better one.

You ask

^_^ Every AI gives you sugar

>_< Bitter finds the flavour

Real output from our early engine, run on Gemma 4 12B, a small open model. People haven't voted on these yet. T2R-001 is being trained to go further.

The proof

In a blind test, of founders picked T2R‑001's ideas over ChatGPT's and Claude's.

We haven't run it yet. The real numbers go here, even if we lose.

  1. 1All three get the same question.
  2. 2Every label comes off.
  3. 3Founders pick what they'd spend a year on.
BenchmarkGPTClaudeGeminiT2R-001

A dash means not measured yet. We only publish numbers we've measured.

Bitter product.
Sweet price.

Start free. Pay when you want more.

Taste

$0
  • 3 ideas a month
  • Checks against the real world
  • The evidence behind each check
Start free

Founder

$9 / month
  • 100 ideas a month
  • Test your own ideas too
  • See what AI keeps suggesting in your market
Get Founder

Team

$49 / month
  • 1,000 ideas a month
  • 5 seats, shared history
  • API access
Get Team

Early pricing. It may change once we've measured real costs.

Founder video coming soon

For two days we asked the best AI in the world for a startup idea. It gave us over 30 and called them great.

Almost all of them were already built, already free in ChatGPT, or wanted by nobody. AI can generate. It has no taste. So we're building the taste.

Rayan Sameer and Wasif Waseem, founders

give me the best yc idea

ngl i think bitter

Yes, we wrote this one ourselves. Even we can't be unbiased about everything.

Be first to taste it.

T2R-001 opens to a small group first. Get in before everyone else has the same idea.

Your tasting ticket

We only use your email to tell you when T2R-001 opens.