NYU's Tal Linzen cites two papers arguing tool use breaks Bender & Koller's 'no meaning' case

tallinzen · x · 2026-09-27

In a debate over the 'stochastic parrot' argument, computational linguist Tal Linzen says he disagreed with both Bender papers when they came out and points to two rebuttals he co-authored: Merrill, Warstadt & Linzen (2022), Entailment semantics can be extracted from an ideal language model, and Mandelkern & Linzen (2024), Do language models' words refer?. His key point: Bender & Koller's argument would not apply to a model that can make tool calls, since tools obviously ground its answers in real-world information.

Original post →

More from AGI Musings

AGI Musings channel →