✳flâneur — a map of the web's best reading

Aether July 2025 Update — LessWrong

lesswrong.com · 1,960 words · saved by 1 readers

Aether is an independent LLM agent safety research group that was announced last August, and a lot has changed in the past 10 months. Now that we’re funded and have officially kicked off, we’d like to share some information about how to get involved, our research, and our team! Aether's goal is to conduct technical research that yields valuable insights into the risks and opportunities that LLM agents present for AI safety. We believe that LLM agents have substantial implications for AI alignment priorities by enabling a natural language alignment paradigm—one where agents can receive goals via natural language instructions, engage in explicit system 2 reasoning about safety specifications, and have their reasoning processes monitored by other LLMs. Within this paradigm, we believe chain-of-thought (CoT) monitorability is a key problem to focus on. There are three reasons why we think that this is a high priority for enhancing LLM agent safety: We are not unique in focusing on monitora

x Aether July 2025 Update — LessWrong Chain-of-Thought Alignment Organization Updates AI Personal Blog 26 Aether July 2025 Update by RohanS , Rauno Arike , Shubhorup Biswas 1st Jul 2025 3 min read 7 26 (Edit: As of Dec 2025, Aether is hiring researchers! You can apply here until Jan 3rd, 2026.) Aether is an independent LLM agent safety research group that was announced last August, and a lot has changed in the past 10 months. Now that we’re funded and have officially kicked off, we’d like to share some information about how to get involved, our research, and our team! Get Involved! Submit a sh

Explore this link on the map →

related reading