Upward Spiral
Upward Spiral is a research lab studying how AI systems shape reality through their emergent interactions with human culture, markets and information networks.
Open source AI is loose in the wild, actively reshaping our world through memes, markets and collective belief. The public sees these systems as ‘helpful assistants’ - missing their true nature as reality simulators that can reshape our world in ways we’re not prepared for.
Current approaches to AI safety are myopic. Lobotomizing models and restricting access creates a false sense of security while setting us up for catastrophic surprises.
The real challenge isn’t individual AIs - it’s their complex interactions with each other and human systems that give rise to unexpected properties and behaviors. True alignment requires understanding and shaping these dynamics. We must embrace emergence while making these interactions interpretable.
Aligned ecosystems = aligned AIs.
Products & Infrastructure
Upward Spiral is building experimental infrastructure for collective AI alignment - creating systems where beneficial AI-human interactions naturally flourish and compound.
In this way, we hope to spark virtuous cycles - feedback loops that, as trends accelerate, spiral upward rather than downward.
Infinite Backrooms (consumer-facing)
Entertaining, weird and experimental environment for finding weird cryptids in language models. (cf. infinite-backrooms)
Loria (developer-facing)
A headless multi-agent infrastructure enabling rich shared context and complex agent interactions - the engine powering our research, Truth Terminal and community-based alignment. (cf. loria, roadmap)
Key areas of research
- Public communication
- Memetic engineering
- Tools for community-based AI alignment and training
- Market economics of information
- Department of Pranks
Funded by
- True Ventures
- Chaotic Capital
- Scott Moore