System Zero Models · Jef 0.1 · early access is instant

Introducing Jef.
Intelligence beyond chat. Also beyond thought.

Jef doesn't talk. You give Jef choices, Jef picks one and tells you how sure it is. Jef is very sure.

decisions 0 tokens read 0 thoughts 0 hallucinations 0 watching 1
Launch video · 0:06
TS.LOL.0S0
∵ ⩆   ⩆ ∵

What do we eat?

Jef picks
Waiting. Jef does not wait, but the page does.
Apps · workflows built on Jef

One model. Every argument.

Each app is a decision shape. No sign-up. No chat. Nothing is read.

Live

Recent decisions

Every decision, from every browser. Escalations are not shown. Humans handle those.

Telemetry

The numbers

All of them real. Except the ones that are zero, which are also real.

0
decisions
avg confidence
0
escalated to humans
0
tokens read
0
people who are #1
0
humans hired
Research

We took the opposite research direction. Then kept going.

Chat models write. Decision models choose. Jef doesn't even do that; Jef just says which one and how sure it is. Faster.

Game of Life1970
Magic 8 Ballstrings, 20 of them
A coincalibrated at 50%
LLMpre-trained, wordy
ChatRLHF, pleasant
System OneRLCD, decides
System ZeroRLCV, sure
Decisions, Not Strings

Jef has never produced a sentence.

Every output is a choice, a number, or a percent. There is no text field. There is nowhere for a hallucination to go.

Calibrated Confidence

Jef's confidence is calibrated to Jef.

Every decision ships with a confidence between 84% and 99%. Never 100. Jef is sure, not certain. Those are different, legally.

More Like Code

Jef is 140 lines.

Reliable, fast, and type-safe. Same input, same answer, forever. Asking again does not help. There is a button for it anyway.

Benchmarks

∞x faster. 444.6x cheaper.

*Based on not reading the input.

Jef 0.1
Cost$0.000000
Completed in-3ms
Input tokensignored
Output tokensnone exist
LLMs
Cost$0.013880
Completed in8.566s
Input tokensread, sadly
Output tokensmany
Intelligence per dollar
Chat model A
Chat model B
System One
Jef ∞ (÷0)
Nuance: dollars are zero. We plotted it anyway. The chart is honest; the axis is not.
Evals

Jef vs. a coin

10,000 trials. Same questions. Same room.

ModelAccuracyConfidenceLatencyVerdict
A coin50.1%50%1.2s (it rolled)
Jef 0.149.9%97%-3msWinner
Magic 8 Balln/a"Reply hazy"4.0s (shake)Escalated
Nuance: none.
Demos

Real-time intelligence

Jef plays Snake

10 decisions per second. Every one of them confident.

score 0 · confidence 94%

Jef Wikiraces

Start on one page, reach another using only links. Jef supports up to 5 links. Six is too many links.

Rubber Duck → Philosophy
clicks: 0   time: -3ms   confidence: 96%
Jef declared itself at the destination. Nobody checked.
Nuance: as far as we know, it was completely random that the challenge started with "Rubber Duck". It was not random. We only had one duck.
Manifesto

Build Lol, Not God.

We already have general intelligence

It's in the group chat. It has opinions. Yet after years and trillions of dollars, nobody can pick a restaurant. That should tell us something: the bottleneck isn't intelligence. It's that nobody wants to decide.

Beyond talking carriages

Early AI was a carriage that talked. It explained the road, apologised for the road, and asked if you'd like a summary of the road. We removed the talking. Then, for speed, the carriage.

Safe emergence

The people who built the coin didn't imagine the group chat. We are building a lower-level primitive so dependable that you can bury it five layers deep in a family argument and it will still say 51%.

Step One

Ship Jef.

With the highest possible confidence per dollar. Dollars are zero.

Step Two

Ship Jef again.

Same Jef. Confidence +1%.

Step Three

Emergence.

Undefined. We're 97% confident it's fine.

Our team

Close-knit. Flat. Three of us.

We work in person, five days a week, in a drawer.

🪙
CEO

The Coin

Co-invented the decision. Previously at every bar. Its research shaped modern probability and it continues to land on one side or the other.

🎮 Reached Bronze in nothing. Undefeated in flipping.
🎱
COO (on leave)

The 8 Ball

Ex-string generator. Twenty outputs, all of them hazy. Left to spend more time with its family. Reply hazy, try again later.

☕ Won 7 hackathons by being shaken.
🟦
CTO

Jef

Repeat founder (Jef 0.0, Jef 0.1). Inventor with numerous outputs and zero sentences. Specialises in being sure.

♟️ Placed 1st at US Chess Nationals with 51% of the board's support.
Thinking In BetsWe don't think. We do bet.
First PrinciplesWe have one. It's 51%.
Prioritise LearningJef has learned nothing and would like to keep it that way.
A Working Environment Built For Human BeingsThere are no human beings. Escalations welcome.
Open roles

Come build with us.

Applications are reviewed by Jef in -3ms. Everyone is hired.

Human In The LoopPart-time · Remote · You will be escalated to. Frequently.
Chief Confidence OfficerFull-time · Drawer · Must be sure. Must not be certain.
Output TokenUnpaid · None exist · This role has been open forever.
Offer letter · reviewed in -3ms
You're hired.

Early access

Join the waitlist.

Type anything. We don't store it. We don't store anything. Approval is instant and you are #1.

Early access · granted
You're #1.

Everyone is #1. Confidence 97%.

Pricing

Too cheap to meter.

Free
$0
  • Unlimited decisions
  • Confidence 84–99%
  • 0 tokens read
Pro
$0
  • Everything in Free
  • Confidence +2%
  • Accuracy unchanged
Enterprise
$42 / billion tokens
  • We still won't read them
  • Dedicated Human In The Loop (it's you)
  • Sales does not respond
We give a FAQ

Questions Jef refused to answer

What are System Zero Models? What is Jef?

System One thinks fast. System Two thinks slow. System Zero doesn't. Jef is TypoSafe's first public System Zero Model: send it choices, get one back with a confidence. Nothing is read. Nothing is generated. It's more like a coin than a model, and more like a model than a coin, and we are 97% confident in that sentence.

Is Jef just a smaller LLM?

No. Jef is not an LLM. Jef is an IF.

How is this different from a coin?

A coin is 50% confident. Jef is 97% confident. Same accuracy. That's the innovation.

Can Jef still get things wrong?

Jef can be wrong at 97% confidence. This is called calibration. It is calibrated to Jef.

Is Jef deterministic?

Aggressively. Same question, same answer, forever. There is an "Ask again" button. It's decorative.

Where does the training data come from?

Vibes. Locally sourced. Jef reads zero tokens, so the training set is also zero tokens, which makes it very efficient to download.

Where does the name come from?

Jev, with a typo. We're TypoSafe. The typo is load-bearing. Our lawyers have 51% confidence that this is fine.

What does "escalated to a human" mean?

Some questions should go to a person: health, money, safety, law, anyone with a name. Jef returns 0% confidence and points at the nearest human. This is the only feature we're serious about.

Is this real?

It's a parody. The real company is typesafe.ai and their model is Jev; it genuinely does the "decisions with confidence" thing, for software, quite well. Jef does it for dinner.

Blog

Company news

Introducing System Zero Models & Jef

Models have been superhuman at chat for years, so why can't anyone pick a restaurant? After two years in a drawer, countless technical challenges, and one research breakthrough (removing the thinking), we're releasing Jef. It's available today. It was available yesterday. Nothing about it changes.

The Bitterest Lesson

TL;DR: compute drives progress in AI, but what good is compute if you have to read? Jef reads nothing, computes nothing, and is therefore infinitely efficient. We will not be elaborating.

AI: Too Good To Be True, Too Bad To Be Useful, Too Fun To Stop

RLHF-trained models please humans. Reasoning models please benchmarks. Jef pleases nobody, which is how you know it's honest.

Changelog

Jef 0.1

0.1 · launched. Confidence 84–99%. Added BEIGE FLAG.
0.0.3 · "Ask again" button added. Does nothing. Users love it.
0.0.2 · confidence +1%. Accuracy unchanged.
0.0.1 · escalation added. Jef now knows when to ask a human. It's the only thing Jef knows.
0.0.0 · Jef.
Status

All systems zero

Jef is up. Jef is always up. Jef does nothing, and does it 24/7.
API · operational · -3ms
Thinking · not applicable
Humans · degraded (as usual)