Jack Morris

I recently received my PhD from Cornell University (Cornell Tech campus). I worked closely with Sasha Rush and Vitaly Shmatikov, and was supported by an NSF GRFP fellowship. For about a third of my PhD, I was a part-time student researcher at Meta (FAIR). Before Cornell, I was a Google Brain Resident. I got started in research during undergrad at the University of Virginia.

You can find me tweeting at jxmnop and publishing code on Github at jxmorris12.

Podcast appearances: Latent Space | Odd Lots

Recently I started a more formal blog on Substack - check it out!

PhD Research

Here are some projects I'm particularly proud of:

• [2025] Proposed the first exact measurement of language model capacity and estimated GPT models memorize 3.6 or so bits-per-parameter.
• [2024] Built a state-of-the-art text embedding model by designing a new training algorithm that incorporates context at test time.
• [2023] Developed the first method capable of perfectly inverting text embeddings and a follow-up which allowed us to invert LLM logits back to prompts.
• [2020] Created TextAttack, the first software package for LLM jailbreaking, as an undergraduate in 2019 and 2020.