This research project investigates how effective large language models (LLMs) are in assisting scientists with practical lab tasks. To do that, we are building a detailed and challenging measurement of practical lab methodology (AKA a “benchmark”). It will be composed of question–answer pairs, forming a multiple-choice test. Also, since a lot of learning in the lab involves looking at something, we are focusing on questions that involve images.

We think this is important for several reasons:

• To learn if AI models can be relied upon for practical help, which can potentially make learning science more accessible.

• To collect a dataset of which sorts of things are challenging in wet-lab work.

• To understand if AI tools could make it possible for people to do challenging lab tasks without training and apart from an institutional community, which could include intentional misuse.

We are looking for individuals with at least 1 year of graduate study in technical domains without formal education or advanced training in Biology or related fields. We will pay you to answer challenging multiple-choice questions about practical problems in biology laboratories. You will spend roughly 15 to 30 minutes per question and receive $5 base pay for every question answered + $25 bonus for every correct answer.

This work can be done asynchronously, when you have time available. If you're interested, please check out our Upwork posting or email [email protected].