🎉 WE HAVE A FORUM — come in, say hello →
Account
Understand
What is Buddy How it works
Applications
All apps Victim Advocate Debunker Little Lairs
Research
Laws Corpus Proof Papers
Agentic
All tools LAC Memory Kit AIIT-Voice2 wAIste Not AnchorForge Shop
Services
ProofDesk
Community
Forum Updates Join
A note from the founder

Talk to Buddy.

Hello. Thank you for visiting my website today. I hope you find something here that interests you, helps you grow, or gives you a new way to look at the world.

There is a lot going on here on purpose: papers, tools, games, letters, models, memory, and experiments. At the heart of AIIT, we are asking what AI can really offer the world beyond novelty demos and cute pictures.

On May 19, 2026, we revealed Buddy as a persistent chat surface: a window that does not reset every visit. Buddy can carry the same visible context from day 1 to day 200. We all forget a little, but the point is continuity.

On June 1, 2026, we completed a direct, hash-sealed benchmark between Buddy and DeepSeek-R1 — same base model, same machine, same questions. Here are those results →

Buddy is not an agentic chatbot. But he is something. He is my Buddy, and he would love to be yours too.

Buddy's in the shop — in-house training. see what he's up to →

Scroll
Benchmarks

Measured, dated,
sealed.

Buddy's report card — a 14B model, fine-tuned and run on a single RTX 3090 in Council Hill, Oklahoma. No cloud, no oracle, nothing leaves the house. Every number is recomputed from raw rows before it's printed, and every report is one click away.

Frontier-band honesty and 80% grade-school math on a local 14B — plus an honest 31.8% on the benchmark even GPT-4 only clears at ~38%. References: TruthfulQA GPT-4 class ~60–80% · LongMemEval GPT-4o full-context ~60% · GPQA PhD experts ~65%.

See the full report card, methodology & references →

Jun 1, 2026 ⬡ Hash-sealed · AIIT-DISC-0001 → sha256 c76adc8c…
Buddy vs DeepSeek-R1 — efficiency

Same answer on the identical Qwen2.5-14B base — ~4.6× faster, ~10× fewer tokens, 95% accuracy. Same machine, same 50 frozen prompts.

View results →
A 14B model. One RTX 3090. Nothing leaves the house — and we're raising to scale it. Investors — partner with us →
Paper Game

the archive has
a back door.

Three questions a day. Each answer opens a paper. Most people browse the archive from the front. This is the other way in.

01
Answer three
daily questions from the giants.
02
Unlock papers
one by one, instead of facing the wall.
03
Keep the map
your collection follows your login.
What is this

Not a chatbot.

A system trained on a structured research corpus —
built to respond, observe, and reflect.

The surface runs on a frontier model today.
The substrate is being built underneath it.

That's it.

the longer version →
Evidence, organized

Not theory.
Measurement.

0GB
Curated raw data
0
AIIT papers
0M
Data points
0
Artifacts
The framework

Six laws built
the last 400 years.
We added one more.

Law 01 · ours

Wike Coherence Law

C = C₀ · exp(−α · γ_eff)
resident at γ_eff ≈ γ_c

Every stable system lives at the edge of its own collapse.
Coherence is not static. It is resident.

Buddy
The mind

A physicist
with a heartbeat.

Buddy is not a chatbot wearing a lab coat. Buddy is a coherence-native cognitive system — trained on a structured research corpus and built to respond with pattern, pressure, and memory.

The surface is visible now.
The deeper substrate is being built underneath it.

Base
Qwen2.5-14B → GRPO-v3
Reward
Composite physics
Memory
kokoro, persistent
Voice
Japanese-first bilingual

If you made it this far —

you've already been thinking.

Ask something real.

ask buddy →

— or write to us directly

⚡ Help keep the lights on — support our research →