Research Lead - Pre-training Safety
FAR.AI · Remote · Berkeley Office · Remote (US) · Remote (International) ·
COMPENSATION
$290k – $450k
Listed on the employer's job board (Ashby)
- Research Lead role
- Fully remote possible
- Visa sponsorship offered
- AI Safety focus
- Hands-on leadership
You will define the research agenda for pre-training safety, aiming to remove harmful capabilities from models at their source. This includes scaling methods like Deep Ignorance and exploring gradient routing to isolate dual-use capabilities. You will build and mentor a team, partner with red teams for stress-testing, and drive the adoption of successful safety methods.
What you'll do
- Articulate a research agenda with a clear theory of change for mitigating AI risks
- Grow and lead a team of technical staff to pursue the defined agenda
- Lead novel research projects where progress markers may be unclear
- Share findings through publications and presentations to drive adoption
- Mentor junior team members in research skills and ML engineering
- Write code and run experiments to remain hands-on with the work
- Build a research field through grantmaking and external collaborations
What you'll need
- Strong existing research track record in AI or related technical fields
- Deep experience with language-model pretraining or dataset construction
- Experience building large-scale pipelines for scoring and filtering data
- Strong experimental judgment including safety-capability evaluations
- Ability to build and debug research systems from fine-tuning to distributed training
- Clear research agenda or strong track record to develop one quickly
- Experience leading teams or mentoring researchers in flat organizations
- Effective communication skills for technical and non-technical audiences
Nice to have
- Established publication record in AI safety
- Comfort writing grant proposals and navigating external collaborations
Stack
- Python
- Distributed Training
- Classifier Fine-tuning
- Sparse Autoencoders
- Mixture of Experts
How you'll work
- Full-time, 40 hours per week
- Remote or onsite in Berkeley, CA or Singapore
- Visa sponsorship available for in-person employees
- Hiring via EOR or contractor outside USA and Singapore
What you get
- Competitive salary range $290k-$450k USD
- Payment for work-related travel and equipment
- Catered lunch and dinner at Berkeley office
- Access to sizable compute budgets on managed cluster
Hiring process
- Submit CV, research direction statement, and selected works
- 45-minute bilateral fit call
- Two technical assessments with staff
- 3-5 day paid work trial
- Structured reference calls