If training Gated SAEs [1] gives better loss recovered against L0 curves than Anthropic's new training techniques [2], on the majority of sites in Pythia-2.8B / Gemma-7B / Mistral-7B (whichever actually gets benchmarked), this resolves yes.

Resolves when information is publicly available about the outcome.

I haven't implemented [2] yet so have no insider information, and also I will not trade in this market besides an initial bet.

[1]: https://arxiv.org/abs/2404.16014

[2]: https://transformer-circuits.pub/2024/april-update/index.html#training-saes

RESOLUTION:

https://cdn.openai.com/papers/sparse-autoencoders.pdf basically shows yes