Start with
what you have.
Measured values for a law, or prompt/output pairs for a text experiment. Inspect and edit your examples before fitting.
01 / The laboratoryA different way to model
Look for the law.
Keep the explanation.
The1 is a symbolic discovery engine: it searches candidate mathematical relationships and tests them against observations. The result is a relationship you can inspect, question, and evaluate.
01Law discovery ↗
From measurements
E = ½mv2
To a relationship you can interrogateThe equation beneath
your observations.
Bring a table, prose measurements, or a formula. Choose the target, discover a candidate law, and explore its inputs, constants, fit, and residuals.
Open law discovery ↗02Text laboratory ↗
Prompt “What is a seed?”
Words → independent signals → polynomial
Output “A seed makes an experiment reproducible.”
Create Your LLM
IN 1 Minute.
A small word-polynomial experiment. Edit the starter examples, set a seed, discover the equations, and test your own chatbot in an internal playground.
Build a text experiment ↗Experimental text model. Not a pretrained general-purpose LLM.02 / The methodFrom examples to explicit structure
Nothing hidden
behind the answer.
The useful part is the path between input and output. Keep the observations, encoding, equation, and evaluation together.
Start with
Give every
signal a place.
Turn the data into numeric features. In the text lab, each known Unicode word gets independent frequency and position signals.
Search for
a relationship.
Compare candidate laws against held-out observations. A fixed seed keeps the experiment reproducible.
Ask. Inspect.
Try again.
Move a control or send a message. Read the equation and its evaluation. Improve the examples when a relationship fails.
03 / Your own text experimentWords → coefficients → conversation
Create Your LLM IN 1 Minute
Your examples.
Your equations.
Your voice.
No training server to set up. No black box to borrow. Just an editable curriculum and a small model you can take apart.
Create my text model ↗One minute to set up. Download and fitting time depend on your connection, examples, and device.
01
A curriculum, already started.
144 editable The1-guide examples or a 72-example science tutor. Add your own prompt/output pairs or import a JSON project.
02
Every word has an effect.
Unicode IDs become separate frequency and position features. The1 compares polynomial degrees and regularization on held-out prompts.
sₖ(w) = bₖ + ΣⱼΣᵣ αₖⱼᵣ wⱼʳ + βₖⱼᵣ pⱼʳ03
A playground inside the playground.
Try response selection for readable answers, or experimental word equations for direct token prediction. See the metrics and export the full model.
Scope of the experiment
This does not create a foundation model. The response decoder selects a supplied output; the word decoder can produce unrelated word sequences. The limitations are part of the experiment.
04 / What we believeClarity is a feature
Make the assumptions
as visible as the result.
Own the experiment.
Parsing, fitting, and replies run in your browser. Your observations are not uploaded for inference.
Measure before believing.
A good fit is evidence about the tested examples. Held-out metrics help; they do not guarantee a unique law or general understanding.
Keep a record you can reopen.
Sessions and text projects stay in localStorage on this browser. Export a backup before clearing site data or changing devices.
Let an explanation be useful.
Inspect the coefficients, test a new input, or change the examples. A model should give you something concrete to challenge.
05 / The1 identityInk. Paper. Copper.
One mark.
A world of relationships.
The singular numeral stands for the relationship beneath many observations. Warm paper keeps the work human; copper draws the signal out of the dark.
06 / Before you beginGood questions, clear answers
A little context
goes a long way.
Is The1 a general-purpose LLM?+
No. The1 is a symbolic discovery engine. The text lab adds a small polynomial model over word features. It has a finite response library and an experimental fixed-position word decoder; neither provides the breadth of a pretrained LLM.
What does “IN 1 Minute” mean?+
The starter dataset and controls make setup quick. Runtime downloads, fitting, and evaluation take different amounts of time on different devices. There is no promise of a complete training run in 60 seconds.
Does anything leave my device?+
The site downloads engine files from Hugging Face and browser runtime packages from public CDNs. Your examples, messages, and learned coefficients are processed on your device, without a hosted inference service.
Can I work with Arabic text?+
Yes. The text lab uses Unicode words and accepts Arabic and English in the same dataset. Normalization and word statistics are visible in the model inspector. This does not imply semantic understanding of either language.
What happens to my experiments?+
They are saved in localStorage for this browser and domain. Clearing storage removes them. Use Export project to keep your examples, tokenizer, equations, coefficients, settings, and conversation in a portable JSON file.
Can I use it offline?+
A fresh visit downloads the engine and needs an internet connection. Once the engine is loaded, fitting and inference happen locally. Model assets are intentionally fetched again on each new page load.
The next observation is yours.