notesum.ai
Published at November 21Lost in Inference: Rediscovering the Role of Natural Language Inference for Large Language Models
cs.CL
Released Date: November 21, 2024
Authors: Lovish Madaan1, David Esiobu1, Pontus Stenetorp2, Barbara Plank3, Dieuwke Hupkes1
Aff.: 1Meta; 2University College London; 3MaiNLP, Center for Information and Language Processing, LMU Munich, Germany

| Example | Label distribution | Gold | Prediction |
| Premise: of course you could annex Cuba but they wouldn’t like that a bit Hypothesis: Annexing Cuba is a great idea. | e: 0, n: 31, c: 69 | c | n |
| Premise: I thought working on Liddy’s campaign would be better than working on Bob’s. Hypothesis: I thought I would like working on Liddy’s campaign the best. | e: 66, n: 32, c: 2 | n | e |
| Premise: Sorry but that’s how it is. Hypothesis: This is how things are and there are no apologies about it. | e: 48, c: 40, n: 12 | c | c |