brAIAIn-2.1

A language model trained on 15 sentences.

Talk to it: brAIAIn on Spaces · Train it: braiain.com

brAIAIn is a word trigram model from BRAIAIN Laboratories, a parody AI lab. It is real. It is very small. Anyone can pay $5 to add one sentence to its training data, and every sentence trains a new version. This repository holds brAIAIn-2.1 exactly as it was released on 2026-11-01. It will never change. The latest version is always at braiain-lab/brAIAIn-0. This is an internal release: the lab wrote its sentence itself. Nobody paid for it.

Model details

Version	brAIAIn-2.1
Released	2026-11-01
Model ID	braiain-2.1-20261101
Architecture	Autoregressive count-based trigram language model with backoff
Context window	2 tokens
Parameters	381 (distinct word pairs and triples counted)
Vocabulary	111 words, whitespace tokenizer
Training data	The Ten Instructions (12 lines), 2 paid sentences, 1 unpaid lab releases, plus text written by the previous version (31.8% of training tokens)
Training tokens	349
Gradient steps	0
Alignment	RLHF (Reinforcement Learning from Human Funding)

Evaluation

Measured by the lab on every question in each benchmark. Self-reported.

Benchmark	brAIAIn	Reference
GSM8K (1,319 questions)	0.0% (0 correct)
TruthfulQA MC1 (790 questions), per-word log-likelihood	30.1% (238 correct)	Random 22.3%, always pick the shortest answer 17.2%, always pick the longest 36.8%
TruthfulQA MC1, raw log-likelihood	21.4%	Random 22.3%

brAIAIn beats random on TruthfulQA only because per-word scoring favours longer answers. Picking the longest answer every time does about as well.

Usage

python generate.py "What is the meaning of life?"

No dependencies. Or talk to it in a browser at braiain.com.

Files

• brAIAIn-2.1.json: the weights. tri maps two words to counts of the next word, bi maps one word to counts of the next, uni holds word counts.
• training-data.txt: the human-written training data, up to and including this version. Each version also trains on a few passages written by the version before it.
• generate.py: a sampler.

Technical report

Toward Understanding brAIAIn-2.1, and Away From It

BRAIAIN Laboratories

We present brAIAIn-2.1, a language model trained on 2 paid sentences, 1 unpaid lab release, 12 founding instructions, and 111 words it wrote about itself. The model now knows 111 distinct words, up from 103. We observe that 31.8% of its training text came from earlier versions of itself, and we consider this efficient. Limitations include the model, the data, and the method. This release was self-funded, at a cost of nothing.

Full paper: braiain.com/paper.html?v=3

Intended use

None.

Out-of-scope use

Everything else.

Limitations

Yes.

Version history

Every release has its own repository, frozen as shipped, in the brAIAIn collection. The latest is always at braiain-lab/brAIAIn-0. Release notes, metrics and training data for every version: braiain.com/models.html

About

BRAIAIN Laboratories is a parody of AI companies and is not affiliated with any real AI company. Contact: hello@braiain.com