Machine Learning Research Engineer

Key Skills

Aartificial intelligence, C++, Effective communication skills, GPUs, JAX framework, Kubernetes-K8s, Large data sets, Machine learning techniques, NVIDIA Jetson, PyTorch

Job Description

Obelisk is the Artificial General Intelligence (AGI) lab at Astera. Obelisk’s mission is to produce AGI in a safe, socially beneficial way. We are focusing on different problems and different approaches than some other AGI efforts. In particular we are focused on the following problems: How does an agent continuously adapt to a changing environment and incorporate new information? In a complicated stochastic environment with sparse rewards, how does an agent associate rewards with the correct set of actions that led to those rewards? How does higher level planning arise?

Our approaches to solving the problems are heavily inspired by cognitive science. To measure our progress, we are implementing reinforcement learning tasks where humans currently do much better than state-of-the-art AI.

Your Mission:

We are looking for a machine learning research engineer to implement architectures, develop environments, train models, optimize performance, summarize results, and help guide research directions.

Location:

You will be required to be in the office in Berkeley at least 2 days per week because we find it significantly improves team cohesion and makes collaboration easier.

Your Responsibilities

• Work with engineers and research scientists to run reinforcement learning experiments to evaluate research ideas.
• Train state-of-the-art models as performance baselines for our research.
• Help design architectures, training algorithms, and training and test tasks.
• Employ rigorous methodology to ensure experiments are valid and reproducible.
• Use all the tools available, including static analysis and unit testing, to write bug-free code.

Basic Qualifications

• 3+ years relevant professional experience, or a PhD in a related field
• Strong programming skills.
• Experience with training reinforcement learning models on new data sets or environments.
• Familiarity with machine learning theory.
• Excited about the potential of AGI.
• Able to communicate technical details of one’s work clearly.
• Able to disagree without being disagreeable.
• Enthusiastic for working on a small, growing team in a low-bureaucracy environment.

Preferred Qualifications

• Experience writing software that is used for a long time or by many people (AKA production-quality code).
• Experience with PyTorch and / or Jax.
• Experience with Ray or other distributed model training systems.

Our tech stack

• We're building environments on top of Minetest, an open source game engine implemented in C++, with a Lua API for building games.
• Our models are written using PyTorch (though we're open to using Jax).
• We have our own cluster of computers with Nvidia GPUs, managed by Kubernetes.

Related roles

• Machine Learning Research Scientist - more emphasis on generating research hypothesis and directions, requires extensive AI theory knowledge.

Why work here?

• Plenty of funding and computers.
• Trying to advance the state of the art in AI.
• Smaller focus. Other places (e.g., DeepMind) are doing research into lots of problems simultaneously, or are doing research and building products (e.g. Anthropic). We are completely focused on a small set of problems.
• Smaller team. This has benefits and disadvantages, but a huge advantage is less communication overhead and bureaucracy. This makes work faster and more fun.
• No outside funding means there’s no pressure to chase trends or make products.