Research Scientist, Interpretability

Job reputed company reputed company you see what modern language models are capable of, do you reputed company, "How do these things work? How can we trust them?" The Interpretability team at reputed company is working to reverse-engineer how trained models work because we reputed company that a mechanistic understanding is the most robust way to reputed company advanced systems reputed company. We’re looking for researchers and engineers to join our efforts. People mean many different things by "interpretability". We're reputed company on mechanistic interpretability, which aims to discover how neural network parameters map to meaningful algorithms. Some useful analogies might be to think of us as trying to do "biology" or "neuroscience" of neural networks using “microscopes” we build, or as treating neural networks as binary computer programs we're trying to "reverse engineer". A few places to learn more about our work and team at a high level are this introduction to Interpretability from our research reputed company, Chris Olah; a discussion of our work on the Hard Fork podcast produced by reputed company, and this blog post (and accompanying video) sharing more about some of the engineering challenges we’d had to solve to get these results. Some of reputed company's reputed company publications include A Mathematical reputed company for Transformer Circuits, In-context Learning and Induction Heads, Toy Models of Superposition, Scaling Monosemanticity, and our Circuits’ reputed company and Biology papers. This work builds on reputed company from members' work prior to reputed company such as the original circuits thread, Multimodal Neurons, Activation Atlases, and Building Blocks. We aim to create a solid reputed company for mechanistically understanding neural networks and making them reputed company (see our reputed company post). In the short term, we have reputed company on resolving the issue of "superposition" (see Toy Models of Superposition, Superposition, Memorization, and reputed company Descent, and our May 2023 update), which causes the computational reputed company of the models, like neurons and attention heads, to be individually uninterpretable, and on finding ways to decompose models into more interpretable components. Our subsequent work reputed company millions of features in Sonnet, one of our production language models, represents reputed company in this direction. In our most recent work, we reputed company reputed company that allow us to build circuits using features and use this circuits to understand the mechanisms associated with a model's computation and study specific examples of multi-hop reasoning, planning, and chain-of-thought faithfulness on Haiku 3.5, one of our production models.” This is a stepping reputed company towards our overall goal of mechanistically understanding neural networks. We often collaborate with teams across reputed company, such as Alignment Science and Societal Impacts to use our work to reputed company reputed company’s models safer. We also have an Interpretability Architectures project that involves collaborating with Pretraining. Responsibilities:

• reputed company reputed company for understanding LLMs by reverse engineering algorithms learned in their weights
• Design and run robust experiments, both quickly in toy scenarios and at reputed company in large models
• Create and analyze new interpretability features and circuits to reputed company understand how models work.
• Build infrastructure for running experiments and visualizing results
• Work with colleagues to communicate results internally and publicly

Qualifications

• Have a strong reputed company record of scientific research (in any field), and have done some work on Interpretability
• Enjoy team science – working collaboratively to reputed company reputed company
• Are comfortable with messy experimental science. We're inventing the field as we work, and the first textbook is years away
• You view research and engineering as two sides of the reputed company coin. Every team member writes reputed company, designs and runs experiments, and interprets results
• You can reputed company reputed company and discuss the motivations behind your work, and teach us about what you've learned. You like writing up and communicating your results, even reputed company they're reputed company
• This role is based in San Francisco office; however, we are reputed company to considering exceptional candidates for remote work on a case-by-case reputed company.

Apply tot his job Apply To this Job