An AI safety research initiative developing new adaptive theoretical frameworks and AI interface designs to keep human sensemaking at pace with rapidly advancing AI systems.
An AI safety research initiative developing new adaptive theoretical frameworks and AI interface designs to keep human sensemaking at pace with rapidly advancing AI systems.
People
Updated 05/18/26lead researcher, Live Theory agenda
developer (Soloware Platform)
researcher and interface developer (Live Conversational Threads, MoSSAIC)
field-building and residency organiser
researcher (Vibe Decoding interface)
project lead (Autostructures) and interface designer
Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- -
Org Details
Updated 05/18/26Live Theory is an AI safety research initiative founded by Sahil Kulshrestha that focuses on developing new theoretical frameworks and AI interface designs to address existential risks from advanced AI. The project operates under the Groundless Alignment organizational umbrella at groundless.ai and is fiscally sponsored by Epistea, z.s., a Czech service organization that supports projects related to long-term survival and flourishing. The core thesis of Live Theory is that traditional formal, systematic approaches to theory and research are insufficient for keeping pace with rapidly advancing AI capabilities. Instead, Live Theory proposes using AI itself as adaptive infrastructure (termed 'autostructuring') that can meet individual intellectual contributions on their own terms, rather than forcing human subtlety to conform to fixed shared structures. This approach draws on concepts from post-formal reasoning and emphasizes context-sensitivity over abstract formulas. The initiative has evolved through several phases. The community originally explored ideas under the label 'High Actuation Spaces,' then developed the Live Theory framework, followed by the Autostructures project at AI Safety Camp (AISC), and more recently consolidated under the Groundless brand. The Live Theory sequence was published on LessWrong and the AI Alignment Forum starting in mid-2024, with key posts including 'Taking Intelligence Seriously' and explorations of interface design philosophy. Practical work includes three main interface projects: Live Conversational Threads (built by Aditya Adiga) for navigating ongoing conversations in real time; Vibe Decoding (by Jayson Amati) for assisting in fine-grained discernment; and the Soloware Platform (by Aayush Kucheria and Kuil Schoneveld) for sharing views on text. A Live Machinery workshop was held at the EA Hotel in November 2024. In 2025-2026, Groundless partnered with AI Safety Camp to run research tracks including Autostructures, AI 2037 (forecasting AI integrity challenges), and MoSSAIC (Management of Substrate-Sensitive AI Capabilities). The initiative has received funding from multiple sources including the Survival and Flourishing Fund, the Long-Term Future Fund, GoodForever, Manifund, and individual donors. Sahil Kulshrestha has prior experience with SERI MATS and MIRI's Agent Foundations team.
Theory of Change
Updated 05/18/26Live Theory posits that as AI systems become more adaptive and capable, traditional systematic and formal approaches to understanding and governing them will fail because they cannot handle context-dependent problems. The initiative aims to develop 'live theories' -- adaptive theoretical frameworks powered by AI that can scale insights without requiring rigid common structure. By redesigning the infrastructure of research methodology and interface design to take advantage of AI assistance, the project seeks to keep human sensemaking at pace with AI development. This includes building interfaces that support postformal reasoning for AI safety, enabling researchers and communities to better understand, anticipate, and respond to risks from increasingly adaptive AI systems. The theory of change runs from better sensemaking tools to improved threat modeling to more effective AI safety research and governance.
Grants Received
Updated 05/18/26Projects
Updated 05/18/26
A Groundless-linked project explored at AI Safety Camp that develops Live Theory–inspired research methodologies and "live interfaces" which use AI to move beyond fixed formal structures toward fluid, context-sensitive sensemaking.
An interface developed by Aditya Adiga that uses AI to identify and manage conversational threads in real time, helping researchers mark and port insights from ongoing discussions into personalised, context-sensitive structures for Live Theory–style sensemaking.
A conceptual AI safety project co-authored by Sahil, Aditya Prasad, Aditya Adiga, and Jayson Amati that analyses substrate-sensitive AI risk scenarios and proposes management strategies, presented as part of the broader Live Theory agenda.
Discussion
No comments yet. Be the first to share your thoughts.