Infinity Intelligence
WillMe GPT
Our public-facing research line: efficient linear RNNs used to test context scaling and training techniques in the wild.
Overview
WillMe GPT is the model line that runs on WillMe AI, our public-facing platform. Where Infinity Intelligence stays strictly for high-reliability institutional use, WillMe AI is where experimental architectures get tested with real users.
It is one of the two model programs that continue after the frontier wind-down, and it is versioned independently of the rest of the division. The current release is WillMe GPT 3.2; earlier and later versions share the same architecture and the same page.
Why linear recurrence
Attention costs grow quadratically with sequence length, which is the single largest reason long-context models are expensive to serve. Linear recurrent models trade some of the flexibility of attention for cost that grows linearly instead.
That trade is exactly the kind of thing that is hard to evaluate in a lab and easy to evaluate with users, which is why this line lives on the public platform.
Where to use it
WillMe GPT is available through WillMe AI, which always runs the current version. It is an experimental model line on an experimental platform, and it does not carry the reliability guarantees that the institutional Infinity Intelligence models were built around.