Jake Levi

[ About · Papers · Talks · Social · Blog · Personal ]

I'm a PhD student on the AIMS CDT supervised by Professor Mark van der Wilk in the Department of Computer Science at the University of Oxford.

I work on fundamental AI research, trying to develop AI systems that can learn more efficiently than current mainstream approaches.

While frontier LLMs have recently achieved staggering success in many domains, their training procedures are expensive, requiring hundreds of millions of dollars, significant environmental cost, and about 4 orders of magnitude more training data than a human requires to master language, every time a model is trained. Even after such great expense, and despite appearing highly intelligent, such models can still be fooled by simple tricks.

I think it is possible that similar performance can be achieved much more efficiently, but not by building on existing foundations (transformer-based architectures, backpropagation-based learning rules, etc). I am trying to develop AI systems that learn more efficiently by researching alternative learning rules and inductive biases. I would consider my current line of research complete when there exists a learning algorithm that can not only demonstrate human-level intelligence, but can also learn from scratch as efficiently as a human. After that, who knows? Maybe AI for healthcare. Or maybe I'll make musical instruments out of carrots.

I believe that the efficiency with which we learn is a fundamental component of natural intelligence. If we don't understand learning efficiency, then we don't understand intelligence (we don't).

Papers

Talks

Social

Blog

Personal

Besides rock climbing (trad & sport & bouldering), cinema, and playing classical guitar, my major passion is designing HTML websites. If you would like me to design your website, email me at
<a href="mailto:example@email.com">ERROR! BROKEN LINK</a>.