Frontier Language Models are not Robust to Adversarial Arithmetic, or "What do I need to say so you agree 2+2=5?
C. Daniel FreemanLaura CulpAaron ParisiMaxwell L. BileschiGamaleldin F. ElsayedAlex RizkowskyIsabelle SimpsonAlex AlemiAzade NovaBen AdlamBernd BohnetGaurav MishraHanie SedghiIgor MordatchIzzeddin GurJaehoon LeeJohn D. Co-ReyesJeffrey PenningtonKelvin XuKevin SwerskyKshiteej MahajanLechao XiaoRosanne LiuSimon KornblithNoah ConstantPeter J. LiuRoman NovakYundi QianNoah FiedelJascha Sohl-DicksteinPublished in: CoRR (2023)
Keyphrases
- language model
- language modeling
- document retrieval
- probabilistic model
- speech recognition
- n gram
- query expansion
- test collection
- language modelling
- retrieval model
- statistical language models
- language model for information retrieval
- information retrieval
- relevance model
- smoothing methods
- language models for information retrieval
- context sensitive
- okapi bm
- passage retrieval
- document length
- query terms
- ad hoc information retrieval
- vector space model
- context dependent
- pseudo relevance feedback
- document ranking