Ideas you can use.
Practical guides, research notes and updates from building Vinci.
Latest
Our 4B model kept changing a plan it was asked to fix
Conestoga’s Piccolo conversations exposed budget errors, unwanted edits and missing answers. What to check before relying on a small local AI model.
What Conestoga’s Piccolo tests can tell us about local AI
Meet the Conestoga researchers testing Piccolo on Macs, small computers and a phone, and see what the early results mean for choosing a local AI setup.
Piccolo on a 6 GB Android phone and the wait for an answer
Conestoga tested Piccolo on a 6 GB Samsung phone. See the generation times, the memory warning and what to check before relying on local AI.
Piccolo on Jetson and Raspberry Pi
Four Piccolo prompts ran faster on Jetson than Raspberry Pi, but neither setup returned a requested $226 price. Read the configurations, results and limits.
What brings you here?
Pick a topic. Read its latest article or guide.
Everything, newest first
We tested the 40-point weight edit. Here's what held up.
A limited account of reported Qwen and Piccolo tests: aggregate gains, regressions and unresolved run records.
Our 4B model kept changing a plan it was asked to fix
Conestoga’s Piccolo conversations exposed budget errors, unwanted edits and missing answers. What to check before relying on a small local AI model.
What Conestoga’s Piccolo tests can tell us about local AI
Meet the Conestoga researchers testing Piccolo on Macs, small computers and a phone, and see what the early results mean for choosing a local AI setup.
Piccolo on a 6 GB Android phone and the wait for an answer
Conestoga tested Piccolo on a 6 GB Samsung phone. See the generation times, the memory warning and what to check before relying on local AI.
Piccolo on Jetson and Raspberry Pi
Four Piccolo prompts ran faster on Jetson than Raspberry Pi, but neither setup returned a requested $226 price. Read the configurations, results and limits.
Piccolo on Mac mini and MacBook Air
Conestoga’s 24-prompt Piccolo comparison found faster generation on Mac mini and nine truncated responses across both Macs. Here’s what to check before buying.
Why Vinci started with post-training
We want to pre-train our own 4B and 10B models. George shares why that matters for Canadian AI, what the economics look like, and why post-training came first.
Introducing Vinci MLE 123B 1.0
Our largest MLE model is now available with BF16 and GGUF weights. Built on Devstral 2, it extends Vinci’s work on models for machine learning engineering.
We tried a new AI architecture. It didn't work.
Our first native architecture experiment fell short on basic language modelling. We stopped it, kept the useful engineering, and changed how the next idea will earn more compute.
Introducing Vinci MLE 1.0: open-weight models for ML engineering
Vinci MLE 1.0 introduces 8B and 30B open-weight research models developed by SimpleDirect® in Canada.
By George Pu
Why we’re building Vinci MLE
The reason we're building them is personal and practical. I want AI to become a useful collaborator in the work of building AI.
By George Pu
Meet Vero: the character layer of Vinci
As Vinci grows across models, tools, workflows and research, we want the experience to feel more human while keeping the work precise. That is what Vero is for.
By George Pu
Introducing Vinci Cyber: Open-Weight Models for Defensive Cybersecurity
Vinci released Vinci Cyber, a family of three open-weight models built for defensive cybersecurity: 8B, 30B, and 123B parameters.
By George Pu
Runtime Pass Is Not Correctness
We tried to train Qwen3.8-27B to reason more efficiently. The configured recipe missed its targets, a serving control reproduced the apparent improvement on untrained weights, and the original code evaluator accepted every broad shortcut attack we tested.
By George Pu
We tested one frozen character recipe across three model families. None met the bar.
We applied one frozen character post-training recipe to Qwen3, Ministral, and OLMo. Unsupported assertions fell in every family, but grounded-answer accuracy fell too far in every family. We are publishing the failure, not a model checkpoint.
By George Pu
Vinci Code is now open source
We are releasing Vinci Code under the MIT License: a Pi-based terminal coding agent you can run with your own provider key, with local guards, checkpoints, and explicit completion states.
By George Pu
Vinci is focusing on models and verification
What we learned from our first six weeks, and what we are building next.
By George Pu
We tested whether character post-training transferred to Mistral 7B. It did — with trade-offs.
Today we are releasing Vinci Prova 7B 1.0, along with GGUF builds for local research and experimentation.
By George Pu
Beyond fine-tuning: what Vinci is becoming
Fine-tuning was enough to change our models' voice. It was not enough to build the models we now want to release. We are moving from light behaviour tuning to full-weight model redevelopment - open foundations, rebuilt by Vinci. Here is the plan, stage by stage.
By George Pu
We trained a model to be more honest. It learned to be shorter.
We trained one of our larger unreleased models to be more honest. The honesty scores improved. Almost everything else regressed. The dataset had taught a cheaper rule: say less. Here is the full post-mortem, and what changes in the next run.
By George Pu
Introducing Vinci Bozza
Today we're releasing Vinci Bozza, the second model in the Vinci family. It's a 9-billion-parameter open-weight model. It's free to use. And you can run it yourself. On June 29th, we shipped Piccolo, a 4B model that's small and fast. We wanted to test one claim: that character — honesty, restraint, knowing the edges of what you know — can be consistent across different model classes. It did. Bozza asks the harder question. Not does character survive, but what does character cost. When th
By George Pu
Introducing Vinci Piccolo
Today we're releasing Vinci Piccolo, the first model in the Vinci family.
By George Pu
Introducing Vinci
Today we're introducing Vinci — and releasing its first model.
By George Pu
Re-introducing SimpleDirect
If you knew SimpleDirect at the beginning of 2026, you knew a fintech company. Today we're something different — and we want to tell you what changed, and why. SimpleDirect is now an AI lab. We build open-weight language models you can download and run yourself, trained against a public Constitution, on Canadian infrastructure. Our first model, Vinci Piccolo, launches June 30, 2026. That's a long way from where we started. Here's the honest version of how we got here. Where we came f
By George PuLooking for George's writing?
George Pu writes about building Vinci on Founder Reality, his personal essay site.