About

Toward Human-level Tutoring with AI

Trillions USD are being spent teaching machines to improve machines.
StudentBench shows the value of teaching machines to improve humans.

What we found: This study establishes that, in a one-hour GRE session, AI tutoring improved students' unaided performance and produced learning gains equivalent to human tutoring overall.

Does StudentBench suggest AI should replace teachers?

Strongly no. The emotional and retention benefits of learning with teachers and peers cannot be overstated. We believe sustained learning and growth with other people matter more to a student's development than what StudentBench provides.

Our mission is access: students should not need the most expensive model or tutor to prepare for the GRE and pursue a better life.

Why we created StudentBench

I grew up in rural Kentucky, about 45 minutes outside Lexington, in a stone house shared by a family of five. I was raised by a mail carrier; my grandfather and great-grandfather carried mail as well. That upbringing shaped a belief that talent is broadly distributed, but access to opportunity is not.

Projects like StudentBench are in the DNA of my work. I built Cleanlab, used by half of Fortune 500 companies, and invented the earliest cheating detection algorithms for online courses on edX and Coursera.

The question that guides my work, including StudentBench, is:

“How do we build a future where AI works reliably in the real world, and in its use expands human agency and distributes opportunity more broadly — rather than concentrating it?”

Inaara Hasmani was my operations counterpart in making StudentBench happen. I couldn't have done it without her.

Ops, Eng, Data Team

• Inaara Hasmani
  Project & Operations Lead
• Trevor Khangi
  Operations Lead
• Justin Pyne
  Forward Deployed Engineer

Research team

• Andreas Plesner
  Research Scientist
• Kevin Feng
  Research Scientist
• Jonas Mueller
  Research Director, Benchmarks

Handshake AI Research

StudentBench is created and managed by Handshake AI Research. The project combines work with frontier AI models, human data at scale over thousands of students, and research at the frontier of AI and human intelligence augmentation.

You can view more benchmarks from Handshake AI Research here.