Evaluating AI capabilities
Evals tell us what AI can do, but shouldn't be taken at face value
TorchbearerCommunity
Humanity has a coordination problem. Torchbearer Community has a plan for that.
Live Last read · last published · next check
Latest posts
Summer reading list: Introduction to AI safety thought leaders
This season flew by but there's still time to get familiar with key concepts and writings from influential people in the AI safety and policy space.
The growing need for global action on AI policy
Politicians, CEOs, and leading researchers have called for international coordination on AI, but what are the mechanisms that lead to meaningful changes that last?
The end of “practical obscurity”
How AI is rewriting privacy, security, and encryption and what organizations can start doing today to prepare.
Direct Institutional Plan in action: a success story
In this post we’re focusing on one of the progress of Samuel Buteau, an AI safety scientist based in Canada. He is a founding member of Torchbearer Community (TBC) and currently supports ControlAI.
Carrying the weight
Member story from Kathleen about how it feels at times being one of the few people in her friend/family group paying attention to the implications of achieving AGI.
The state of cybersecurity in the AI age
Walls that keep us relatively safe have been breached + steps to patch and reinforce them
Robust to what?
The hidden civilizational invariants we rely on are not guaranteed, especially in the age of AI.
We are likely in an AI overhang. And this is bad.
How unknown capabilities in AI models may undermine most AI safety frameworks.
Could AI engineer a pandemic?
Unfortunately, there's a precedent, and the technology is quickly making this scenario alarmingly plausible.