You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
I have my M.S. in Computer Science (Data Science) and 4 years of startup ML engineering experience. I am a visiting researcher at the University of Edinburgh.
I work in the field of Mechanistic interpretability (explainable AI, understanding model internals) for large language models, and model evaluation. I attended the New England Mechanistic Interpretability (NEMI) conference this year and presented a poster on an evaluation framework I am working on: https://zenodo.org/records/21913952
The workshop I will be attending is: AI Measurement Science: Toward Rigorous AI Evaluation (AIMS) — treats AI measurement as a scientific discipline drawing on machine learning, statistics, psychometrics, economics and policy. https://aimslab.stanford.edu/workshop
My work is a strong fit for this as it draws directly from psychometrics and provides a rigorous framework for measurement validity for mechanistic claims about neural networks. The idea is triangulation of evidence and providing a reportable scorecard to measure things like necessity, sufficiency, minimality, falsifiability, as well as new criteria drawn from fields like neuroscience, genetics, pharmacology, etc.
Anthropic J-space (Gurnee et al.): four different claims at four evidence levels.
- A low-dimensional Jacobian object mediates which contents the model reports and reasons over — Mechanistically Supported
- Any direction encoding the concept would serve equally — Disconfirmed, tested twice, concept vectors don't reproduce the effect
- It is a global workspace in the sense the theory intends — Underdetermined, no experiment separates a workspace from a bottleneck that behaves like one
- The result holds beyond one model family — Insufficient, four models, one lab, all closed-weight
The project has been completed and I am just traveling to the conference to get feedback on the work and show it to a new community (the AI/ML evaluations in general, versus mechinterp community).
Travel expenses for the workshop. The main conference is sold out so it will only be October 9th. I live in Boston, MA so I will be traveling cross-country.
The flight I am planning to take is $397 nonstop. Hotel would be three days for $150-200 a night. Total
I am working solo. I have a published paper at ICML 2022 - https://proceedings.mlr.press/v162/das22a.html and recently prented a poster at the NEMI 2026 workshop.
Here is my personal website: https://elliottower.ai/
N/A - Just need help accommodations-wise.
None.
There are no bids on this project.