MODELS / OPEN CHANNEL
COMPARENOTES.
Training runs, inference behavior, architecture choices, data, evaluation, or the weird thing your model is doing at 2 a.m. This is the place for the details.
Model work gets useful when the black box opens up.
GregRietta AI is being built by understanding the stack instead of hiding it behind an API. Small model experiments, training behavior, inference constraints, and evaluation all belong in the conversation.
The interesting part is usually not whether a model can produce text. It is why it behaves the way it does, what the hardware is actually doing, and what changes when the system is pushed outside the happy path.
WHAT TO BRING / 02
Enough detail to make it interesting.
Bring the behavior.
Loss curves, strange generations, throughput changes, failures, or anything else that made you stop and ask why.
Bring the tradeoff.
Architecture, tokenizer, dataset, parameter count, context length, quantization, and deployment decisions are all fair game.
Bring the thing you cannot explain yet.
A specific unknown is usually more interesting than a polished result. The point is to compare notes and understand the system better.
DIRECT LINE / 03