AI Systems, Language & Reasoning Models

About the role

Unconventional AI is building a new computing foundation designed to increase AI energy efficiency by 1000x by mapping neural networks directly to semiconductor physics. We are seeking a Member of Technical Staff to bridge the gap between NLP research and hardware design. You will build language and reasoning models that operate natively on our unique silicon architecture.

Key facts

What you'll do

• Architect and train advanced language and reasoning models, including transformers, state space models, and diffusion or flow models, optimized for our hardware.
• Develop sequence modeling techniques that utilize the continuous-time dynamics of our silicon rather than traditional digital abstractions.
• Create training recipes, loss functions, and evaluation benchmarks to achieve high-speed reasoning and comprehension while maximizing energy efficiency.
• Work directly with hardware engineers and system builders to align model architecture with physical compute primitives.

Requirements

• MS, PhD, or equivalent research background in Computer Science, AI, Machine Learning, Physics, Electrical Engineering, or Applied Math.
• Practical experience in the theory and training of foundation models such as transformers, SSMs, or text diffusion.
• Proven ability to manage large-scale distributed training runs on significant compute clusters.
• Proficiency in PyTorch or JAX with a history of writing scalable, clean code for large language models.

Nice to have

• Background in hardware-in-the-loop training, mixed-signal hardware, quantization, or physics-informed neural networks.

Skills & tools

• PyTorch
• JAX
• Transformers
• State Space Models
• Diffusion/Flow Models
• Distributed Training
• Hardware-in-the-loop training

Practical notes

Benefits include comprehensive health coverage, 401k matching, and unlimited PTO. Complimentary meals are provided at our Palo Alto office.