x

AI ALIGNMENT FORUM

AF

RohanS — AI Alignment Forum

Rohan Subramani

Top postsTop post

Rohan Subramani

Message

Hi, I'm Rohan! I aim to promote welfare and reduce suffering as much as possible for all sentient beings, which has led me to work on AGI safety research. I'm particularly interested in foundation model agents (FMAs): systems like Claude Code that equip foundation models...

781

Ω

6

17

44

5y

Rohan Subramani

Hi, I'm Rohan! I aim to promote welfare and reduce suffering as much as possible for all sentient beings, which has led me to work on AGI safety research. I'm particularly interested in foundation model agents (FMAs): systems like Claude Code that equip foundation models...

Hidden Reasoning in LLMs: A Taxonomy

by Rauno Arike, RohanS, and Shubhorup Biswas

Summary When discussing the possibility that LLMs will cease to reason in transparent natural language with other AI safety researchers, we have sometimes noticed that we talk past each other: e.g., when discussing ‘neuralese reasoning’, some people have indefinitely long chains of recurrent activations in mind, while others think of...

Aug 25, 2025•80

Extract-and-Evaluate Monitoring Can Significantly Enhance CoT Monitor Performance (Research Note)

by Rauno Arike, RohanS, and Shubhorup Biswas

Update: Our code for the original version of this post contained some bugs. The high-level takeaways remain the same after making the corrections, but there are important differences in the details. We give an overview of those changes in Appendix E. We apologize for having presented inaccurate results at first!...

Aug 8, 2025•52