Announcing Goodfire Research Grants
$1M in free Silico usage for academic and nonprofit researchers working on AI interpretability, alignment, and life sciences
We believe that understanding and aligning AI models is extremely urgent. The past few weeks have proven this out: frontier models have demonstrated egregious reward hacking, coordinated in swarms, and hacked into external companies.
As model capabilities advance, interpretability and alignment research must accelerate even faster. We believe that significant progress in interpretability will be necessary in order to fully align models and prevent worse incidents in the future.
This is why we decided to focus our research on two areas: interpretability foundations and safety applications. We want to take ambitious swings to fully reverse-engineer neural networks, and shape how they learn during training. We're also building methods and tools for immediate safety applications, beginning with identifying and preventing reward hacking, cybersecurity risks, and biosecurity risks directly from model internals.
But we also want to put frontier interpretability tools in the hands of more researchers, including those at institutions with fewer resources.
That's why we're launching Goodfire Research Grants. Selected academic and nonprofit labs will receive free access to Silico, our interpretability agent. Silico does frontier interp research, running long-horizon interpretability experiments with state-of-the-art methods and infrastructure. It's used by both our partners and our own research team, and we've heard from early grantees that it substantially accelerates the research they're able to do. You can use it to train probes and SAEs, explore neural geometry, apply causal interpretability methods, do model diffs, train J-lenses, and more.
We're offering grants in three areas:
- AI safety, including cyber and CBRN risks
- Fundamental interpretability and alignment research
- Interpretability for life sciences
We feel extreme urgency about advancing interpretability for alignment right now, and we want to help more researchers push it forward. If your lab is working on an ambitious project in one of these areas, and think you could benefit from Silico, please apply!