University of Warwick ⋅ PhD Computer Science ⋅ 2022-26 – supervisor: Long-Tran-Thanh
Imperial College London ⋅ Undergraduate Masters (MSci) Math ⋅ 2018-22 – 1st class honors
Goldman Sachs ⋅ AI Research Intern ⋅ Jan-Aug 2021, London – Developed and implemented novel methods in NLP and recurrent neural networks for financial forecasting
My primary work focuses on topics at the intersection of markets and AI, especially AI alignment.
Abhimanyu Pallavi Sudhir and Long-Tran Thanh (2024), “Betting on what is neither verifiable nor falsifiable”, arxiv.org/abs/2402.14021
Abhimanyu Pallavi Sudhir (2021), “A mathematical definition of property rights in a Debreu economy”, arxiv.org/abs/2107.09651
Related write-ups and talks.
Lesswrong post (2024): “Reinforcement learning from market feedback, and other uses of information markets”
LessWrong post (2023): “Betting on what is un-falsifiable and un-verifiable”
Poster at the Co-operative AI Foundation (CAIF) summer workshop, 2023:
LessWrong post (2022): “Meaningful things are those the universe possesses a semantics for”
Developed a Consistency Benchmark for LLM forecasters, along with a principled arbitrage-based metric for inconsistency and a novel consistency calibration method similar to Platt scaling.
Abhimanyu Pallavi Sudhir*, Alejandro Alvarez, Adam Shen, and Daniel Paleka* (2024), “Consistency Checks for Language Model Forecasters” Workshop paper, accepted to: Agentic Markets Workshop at ICML 2024; NextGenAISafety Workshop at ICML 2024; Oxford ELLIS Robust LLMs Workshop 2024
Ongoing collaboration with a team supervised by Arjun Panickssery and Nina Rimsky, to develop a comprehensive benchmark for Scalable Oversight protocols.
Abhimanyu Pallavi Sudhir (2019), “Infinitesimal translations and a multivariate Grünwald-Letnikov calculus”, arxiv.org/abs/1904.02710
Abhimanyu Pallavi Sudhir (2019), “Generalisations of the determinant to interdimensional transformations: a review,” arxiv.org/abs/1904.08097
Abhimanyu Pallavi Sudhir (2018), “The generalized Cauchy derivative as a principal value of the Grünwald-Letnikov fractional derivative for divergent expansions,” arxiv.org/abs/1809.08051
Abhimanyu Pallavi Sudhir (2014), “On the Determinant-like function and the Vector Determinant,” Advances in Applied Clifford Algebras (24-3: 805-807), doi:10.1007/s00006-014-0455-3
Abhimanyu Pallavi Sudhir (2013), “Defining the Determinant-like function for m by n matrices using the exterior algebra,” Advances in Applied Clifford Algebras (23-4: 787-792), doi:10.1007/s00006-013-0416-2
Teaching Assistant for CS255: Artificial Intelligence (Warwick) ⋅ 2024
Reviewer for NextGenAISafety Workshop at ICML 2024 ⋅ 2024
Teaching Assistant for CS141: Functional Programming (Warwick) ⋅ 2023
Reviewer for Advances in Applied Clifford Algebras (Springer) ⋅ 2020
Co-operative AI Foundation ⋅ Jul 2023 ⋅ workshop on AI and cooperative game theory
Wrote the Python package costly for estimating costs and
running times of complex LLM workflows/experiments/pipelines in advance
before spending money, via simulations.
Project page: github.com/abhimanyupallavisudhir/costly
Install: pip install costly
Final-year MSci project with Professor Jeroen Lamb at Imperial College London exploring equivariant learning and causal DAGs.
Report: abhimanyu.io/legacy_writing/Imperial_reports/m4r.pdf
Undergraduate research project with Professor Richard Thomas at Imperial College London on Lie groups and algebras.
Report: abhimanyu.io/legacy_writing/Imperial_reports/urop.pdf
Presentation: abhimanyu.io/legacy_writing/Imperial_presentations/lie_theory.pdf
Computerized formal proving in Lean with Professor Kevin Buzzard at Imperial College London.
Wrote the FilterProduct.lean and Hyperreal.lean modules for the Lean math library
Formalized the first-year “Foundations of Analysis” module exam Blog post:
xenaproject.wordpress.com/2019/05/06/m1f-imperial-undergraduates-and-lean/
Co-founded PhysicsOverflow, a postgraduate-level physics Q&A site and open peer review system. See en.wikipedia.org/wiki/PhysicsOverflow for more details.
Abhimanyu Pallavi Sudhir and Rahel Knoepfel (2015), “PhysicsOverflow: A postgraduate-level physics Q&A site and open peer review system,” Asia-Pacific Physics Newsletter (4-1: 53-55), doi:10.1142/S2251158X15000193
Email: [email protected]
Contact: +44-7771824896
Website: abhimanyu.io
Google Scholar: scholar.google.com/citations?user=lb38BjYAAAAJ
LessWrong: lesswrong.com/users/abhimanyu-pallavi-sudhir