Major Labs Achieve RSI: OpenAI, Deepseek, and the Race to AGI

teortaxesTex · x · 2026-08-16

Argues that every serious lab has implemented Recursive Self-Improvement (RSI). OpenAI has an "AGI Intern" scalable in weeks. Kimi and GLM are doing RSI. Deepseek reportedly has an Opus-level model processing trillions of tokens daily for autoresearch, soon reaching Mythos-Preview levels.

The discussion questions how leaders like Dario Amodei and Sam Altman will decide human fate, noting that human participation doesn't preclude superintelligence and that parallelization is already mastered.

Related event: Anthropic Report Reveals Internal Model 2 Exceeds Mythos 5(21 posts)→

Original post →

More from AGI Musings

AGI Musings channel →