Akshat Gupta

I am a fourth-year PhD Candidate at UC Berkeley, advised by Gopala Anumanchipalli, affiliated with BAIR and Berkeley Speech Group.


Research Goal: "To make parameter modification a normal part of model use."


I am actively working on the problem of Continual Learning. Some of my representative ideas and works can be found below:


As an AI researcher, I'm tired of the model weights being frozen during use. CL is a nontrivial problem which will also introduce new types of safety and security concerns. Hit me up if you want to chat about it.

Email  /  Google Scholar  /  Twitter  /  Linkedin  /  CV (Updated Aug 2026)

profile photo
Experience
Google Student Researcher Apr 2025 – Aug 2025
Automated discovery, multi-agent systems, and code & systems optimization.
IBM Research Research Scientist Intern May 2025 – Aug 2025
Latent reasoning in large language models.
JPMorgan Chase Senior Associate, AI Research Jun 2021 – Aug 2023
Relation extraction, knowledge graphs, and large-scale public information discovery.

Blog

  • I wrote a blog on how we're slowly getting ready for continual learning agents - From Thinking Models to Learning Agents.

  • I recently wrote a blog about looking at Humans from the lens of AI, called "Un-Anthropomorphizing Humans".
    • The blog contains some of my thoughts on continual learning and the future of AI and humans. I'd love to hear thoughts about it!

  • I reviewed one of the AI generated papers by Sakana AI. If you're interest, here's the link.


News