Eltayeb Ahmed
@clockwk7
PhD student at Oxford Uni. Ex: Google Research, FAIR. University of Khartoum and AIMS Rwanda Alumni.
Customers build all IKEA products at home. But this comes with benefits... A "Billy" bookcase uses 6-8x less shipping space disassembled—lower shipping costs keep prices stable for buyers.. But there's another benefit:
Can an LLM be programmed? In our new preprint, we show that LLMs can learn to evaluate programs for a range of inputs by being trained on the program source code alone – a phenomenon we call Programming by Backprop (PBB). 🧵⬇️
How can we bypass the need for online hyper-parameter tuning in offline RL? @FLAIR_Ox is introducing two fully offline algorithms: SOReL, for accurate offline regret approximation, and TOReL, for offline hyper-parameter tuning! arxiv.org/html/2505.2244…
Why am I writing a fully functional algebraic expression parser for my custom hyperparameter tuner 2 days before the deadline to submit my NeurIPS appendix?
12 hours to deadline. I finally give in and add the inevitable \vspace{-*}. Next submission will be the one I don’t do this.
It is a testament to the crazy age we live in that every time I attend a talk by a GPU/TPU design person, they complain about the speed of light being a real limitation for current chip paradigms.
Never ask a man his salary, a woman her age, and a number theorist where the log term came from.