alphaBell

company

https://alphabell.com

alphabell-research

AI & ML interests

Post-training, model behavior and transparent evaluation. Research from alphaBell.

Recent Activity

alphabell-research new activity about 7 hours ago

alphabell/AI-by-the-Numbers:Clamp playback time at the start of a frame

alphabell-research new activity about 7 hours ago

alphabell/AI-by-the-Numbers:Bind output tokens to committed weights

alphabell-research new activity about 19 hours ago

alphabell/AI-by-the-Numbers:Correct memory notes before reusing them

View all activity

Articles

Ling 3.1 Flash 1019: same setup, new weight shards

Draw a MOSS layout once, then run it in ComfyUI

52 full-attention layers versus 16: mapping Nanbeige and Qwen

Five experiments inside Flint and Qwen Open

Draw MOSS-Image layouts without doing coordinate arithmetic

Three tiny bugs that broke Flint's games

alphaBell Qwen Open: 27B post-training, with the tradeoffs measured

View all articles

alphabell 's datasets 1

alphabell/Flint-Arcade-Trajectories

Viewer • Updated 18 days ago • 32 • 96