Each presentation plays slide by slide in a new tab. Use the arrow keys, click, or the on-screen controls, and press F for full screen. Every one is also print-ready to PDF.
Most companies will rent intelligence. A few will build it. A working briefing on building small language models you own, end to end. The build threshold that decides use versus adapt versus build, what goes inside the pipeline, why your data is the moat and synthetic data the multiplier, architecture and pre-training, post-training, evaluation under real budgets, fast inference, what it takes in team and budget and time, when not to build, and the one move to start. Positions the Zorp open toolchain diplomatically.
A confident answer is not a defensible one. A working briefing on turning a hard question into an investigation that keeps a record every claim can be traced to. The gap between fluency and accountability, pre-registration and the kill threshold, the evidence record every claim resolves to, a human gate before each step, and the one move to start with. Answers are cheap. Evidence is not.
Everyone rents intelligence by the token. What if you owned the whole stack? A working briefing on an open stack that runs any model on your own hardware, with memory, tuning, evaluation, fast inference, a research agent, and the security to run it all offline. Run an open model, adapt it with fine-tuning and post-training, or build your own. Everything stays in your walls, and it moves at the speed of open source.
Everyone is talking about AI alignment. What would it actually look like? The working blueprint for governing autonomous agents as they move from pilots into real deployment. Govern the deployment, not the model. A reference architecture, four controls to put in place before an agent ships, a human-oversight rule by reversibility and consequence, and the one move that matters, mapping what each agent can reach against what it was authorized to reach.