Hi Keri,

I completed a 4-year diploma this year and have not started my bachelor's degree. My sole-author paper, "Training Quality Determines Efficiency Boundaries in Test-Time Reasoning," has been accepted as a poster at NeurIPS 2026 in Sydney: https://neurips.cc/virtual/2026/loc/sydney/poster/150721

To be upfront, the paper is not a security paper. It studies how much a reasoning model's training quality, rather than just its size or token budget, determines how reliably it reasons at inference time. Across a broad set of open reasoning models, I found that a simple training-quality score tracks efficiency strongly, and that brevity constraints alone can recover a large share of accuracy lost to truncation. I think this kind of work matters for frontier systems, because it gives a cheap way to tell which models can be trusted to reason reliably before deploying them. I'd be glad to talk about how it connects to evaluation and reliability work in security.

I did this research without a lab or a supervisor. NeurIPS requires an author to present in person. NeurIPS has given me complimentary registration and seven nights of hotel accommodation, but I have no funding for the rest of the trip: the flight, food and local transport. My institute and the education board gave no support, and other organizations I contacted did not respond or turned me down because of my education level.

My fundraiser is here: https://manifund.org/projects/undergrad-sole-author-neurips-2026-main-track-paper-travel-to-present-in-sydney

If you can support it in any amount, or point me to someone who can, I would be sincerely grateful. I can share any proof or details you need.

Google Scholar: https://scholar.google.com/citations?user=jVyIovcAAAAJ
Website: https://logicsame.github.io/md-hakim.github.io/