ML Research Resident

Elicit·Oakland, CA (or remote within US timezones)·Posted 13 December 2024

Apply with a tailored CV View on Elicit's site

About the role

Elicit is building a research agent that can use an unlimited amount of test-time compute while keeping its reasoning transparent and verifiable. THE RESIDENCY Transformers do a fixed amount of computation per token, and the quality of work degrades rapidly when they are applied iteratively. As research resident, you'll work with us for 3 months on developing computational procedures (operators) that can reliably improve a knowledge state over thousands of iterations. What is a knowledge state? A knowledge state consists of structured information - for example, a scientific paper might be represented as a set of claims supported by evidence and connected through logical reasoning; this might be combined with scratchpads, evergreen “notes to self”, search trees, and other information. What counts as improvement? Like scientists, we want LLMs to make genuine progress in understanding - separating inferences from raw evidence, finding connections between ideas, building clearer explanations, and identifying gaps in reasoning. But unlike typical ML systems that are often trained to do “whatever works”, we need improvements that are epistemically sound - each step should make…

This is a summary of the advert. The full description and the application are on Elicit's careers site.

Apply faster with Prepper

A version of your CV tailored to this ML Research Resident role.
A cover letter written from your CV and this advert.
Prepper can fill in Elicit's application form and send it once you approve.
The application saved in your tracker, with reminders to follow up.

Start free

More roles at Elicit

Prepper is independent and not affiliated with Elicit. Roles are listed from the employer's public careers system and removed when they close.