Started as a 240M parameter experiment. Died from a lack of compute. Came back as something better.
240M PARAMSMOELLAMAGPTPOWAHHHHREBORN240M PARAMSMOELLAMAGPTPOWAHHHHREBORN
Evolution.
MoE
Mixture of Experts. The beginning.
LLaMA
Open foundation. Better architecture.
GPT
Generative Pre-trained. Final form.
Live — v2.0
Maxim 2.0
The vision attempt hit a compute wall — not enough POWAHHHH. So 2.0 borrows it: a full AI chat app with accounts, conversation history, and streaming, served from my own infrastructure at maxim.alflora.app.
Talk to Maxim2023–2024: MoE → LLaMA → GPT, trained on coding Q/A. Deprecated. 2026: reborn.