One of the main directions of research we care about at Imbue is developing better theoretical models and understanding of learning, optimization, memory, and agents. In particular, we're interested in the intersection of this theoretical knowledge and what that means for our day-to-day engineering of more general agents. This position is about developing the software to support those experiments and explorations.

• Creating a service to compute convolutional neural tangent kernel matrices for use by both internal and external users

• Investigating the ways in which hyperparameters transfer to larger scale models (ex: extending μTransfer to other models and parameters)

• Creating toy examples, graphs, and visualizations to help understand the learning-time mechanics of self-supervised networks like BYOL, VICReg, and others.

• Running thousands of experiments to empirically determine the optimal learning rate schedule for a network.

• A good programmer. Much of the work will require quickly putting together experiments and one-off visualizations. It will require both attention to detail, and speed.

• Mathematically inclined. Your work will benefit from a broad knowledge of (and comfort with) mathematics. In particular, knowledge of kernels, matrices, information theory, and related topics will be helpful.

• Passionate about science and understanding how deep learning actually works. You should be driven to really deeply understand these topics and be unsatisfied with hand-wavy explanations.

• A good communicator. You will need to communicate your results and questions to both the rest of the team, and the broader world via papers and blog posts, both of which require an ability to explain complex ideas in a succinct, easy-to-understand manner.

• Work directly on creating software with human-like intelligence.

• Generous compensation, equity, and benefits.

• $20K+ yearly budget for self-improvement: coaching, courses, conferences, etc.

• Actively co-create and participate in a positive, intentional team culture.

• Spend time learning, reading papers, and deeply understanding prior work.

• Frequent team events, dinners, off-sites, and hanging out.

All submissions are reviewed by a person, so we encourage you to include notes on why you're interested in working with us. If you have any other work that you can showcase (open source code, side projects, etc.), certainly include it! We know that talent comes from many backgrounds, and we aim to build a team with diverse skillsets that spike strongly in different areas.

We try to reply either way within a week or two at most (usually much sooner).

Imbue builds AI systems that reason and code, enabling AI agents to accomplish larger goals and safely work in the real world. We train our own foundation models optimized for reasoning and prototype agents on top of these models. By using these agents extensively, we gain insights into improving both the capabilities of the underlying models and the interaction design for agents.

We aim to rekindle the dream of the *personal* computer, where computers become truly intelligent tools that empower us, giving us freedom, dignity, and agency to pursue the things we love.