Tag
The article reports initial results from OpenRSI's Marin-Scaling-Ladder experiment, where AI agents autonomously propose, implement, and evaluate new optimizers across model scales from 550M to 2.5B, discovering PSPR with Codex (GPT-5.6).