L
LawZero Tech Stack
AI safety research nonprofit building non-agentic systems and oversight tools
LawZero is a Montreal-based nonprofit conducting AI safety research under the direction of a leading AI researcher. The tech stack—PyTorch, TensorFlow, JAX, distributed training infrastructure (DeepSpeed, Ray, SLURM), and large-scale inference optimization (vLLM, Hugging Face Accelerate)—reflects heavy computational work on model training and inference efficiency. The research-dominant hiring mix (11 of 16 active roles) paired with projects spanning experimentation pipelines, probabilistic graphical models, and amortized inference methods signals a team built for rigorous, reproducible AI safety experiments rather than product delivery.
Tech Stack 16 technologies
Core StackPyTorch TensorFlow AWS Docker Kubernetes Python gRPC JAX GCP Azure DeepSpeed Hugging Face Accelerate vLLM Ray SLURM NVIDIA Nsight
What LawZero Is Building
◆Challenges
- Malicious use of ai
- Large model training challenges
- Advanced ai security problems
- Mitigating ai risks
- Performance bottlenecks in distributed training
- Resource utilization optimization
- Scaling large model inference
- Misalignment risk of llms
- Risks posed by advanced ai models
- Scaling ai talent acquisition
▲Active Projects
- Ai security research agenda
- Novel ai safety agenda
- Large-scale experimentation pipelines
- Research experiments across toy scenarios
- Tools and libraries to optimize computing resources
- Amortized inference methods for high-dimensional discrete and continuous distributions
- Parameter- and structure-learning methods for large probabilistic graphical models
- Experimental protocols and evaluation frameworks
- Scientist ai program
- Scientist ai agenda
Hiring Activity
15 roles · 10 in 30d
Department
Seniority
Company intelligence
