Akira Yoshiyama
Currently working on (1) diffusion language models and (2) test-time training1 for LLMs with cool people. Things I've been a part of: I've lived in SF, Zurich, Waterloo, Vancouver and Montreal. My favorite poem is If. Twitter | Google Scholar | Curius | LinkedIn
About Notes Admirations Photos I'm interested in accelerating research (including AI research) with AI. I'm on a gap year from the University of Waterloo, living at The Residency in NYC. Currently working on (1) diffusion language models and (2) test-time training1 for LLMs with cool people. Things I've helped bring to life and/or maintain: LADDER & TTRL: self-contained curriculum learning techniques for complex problem-solving. Socratica: a wonderful community of creatives, makers, and artists doing cool things together. Accelerating Neural Architecture Search: low overhead model…
saved by
related reading
- Large Language Diffusion Modelsarxiv.org
- GenAI Handbookgenai-handbook.github.io
- machine learning imindslice.substack.com
- 2506.17298arxiv.org
- Mainrush-nlp.com
- NOUS RESEARCHnousresearch.com
- LYDIAlydia.ml
- Kuleshov Group | How to Build a Diffusion Language Modelkuleshov-group.github.io
- Thoughts — Jason Weijasonwei.net
- Neil Rathics.stanford.edu
- Esoteric Language Modelsarxiv.org
- 2025: The year in LLMssimonwillison.net