Back to feed
arXiv cs.CL·

Fodor and Pylyshyn's Systematicity Challenge Still Stands

Signal
72
Hype
15
In three linesAn arXiv paper challenges claims that neural networks solve Fodor and Pylyshyn's systematicity challenge. Authors show Lake and Baroni's meta-learning for compositionality protocol fails on out-of-distribution data and lacks systematicity even within-distribution.
Read source
Your take?
ReasoningBenchmarks

Summary generated by Claude — human-verified