ICLR 2023 accepted papers on Large language models
82 of the 1,575 papers accepted at ICLR 2023 (5.2%) are on Large language models, compared with 2.3% at ICLR 2022. They fall into 6 topics.
See ICLR 2023 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest accepted papers by meaning and how crowded the spot has become.
Topics and papers
Language, safety and interpretability · 46 papers
- GLM-130B: An Open Bilingual Pre-trained Model (293 citations)
- Large Language Models are Human-Level Prompt Engineers (289 citations)
- Quantifying Memorization Across Neural Language Models (162 citations)
- UL2: Unifying Language Learning Paradigms (94 citations)
- Ask Me Anything: A simple strategy for prompting language models (75 citations)
- Prompting GPT-3 To Be Reliable (67 citations)
- Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task (62 citations)
- Selective Annotation Makes Language Models Better Few-Shot Learners (62 citations)
- Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 Small (53 citations)
- Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation (52 citations)
- Language Models are Realistic Tabular Data Generators (48 citations)
- PEER: A Collaborative Language Model (44 citations)
and 34 more in the interactive map.
Agents, code and math · 16 papers
- CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis (231 citations)
- Decomposed Prompting: A Modular Approach for Solving Complex Tasks (99 citations)
- CodeT: Code Generation with Generated Tests (60 citations)
- Binding Language Models in Symbolic Languages (36 citations)
- Generating Sequences by Learning to Self-Correct (30 citations)
- Draft, Sketch, and Prove: Guiding Formal Theorem Provers with Informal Proofs (26 citations)
- Formal Mathematics Statement Curriculum Learning (24 citations)
- DocPrompting: Generating Code by Retrieving the Docs (19 citations)
- Language Models Can Teach Themselves to Program Better (17 citations)
- Planning with Large Language Models for Code Generation (14 citations)
- Summarization Programs: Interpretable Abstractive Summarization with Neural Modular Trees (7 citations)
- Neuro-Symbolic Procedural Planning with Commonsense Prompting (3 citations)
and 4 more in the interactive map.
Retrieval and knowledge · 9 papers
- Generate rather than Retrieve: Large Language Models are Strong Context Generators (85 citations)
- Discovering Latent Knowledge in Language Models Without Supervision (46 citations)
- Promptagator: Few-shot Dense Retrieval From 8 Examples (43 citations)
- Leveraging Large Language Models for Multiple Choice Question Answering (39 citations)
- Recitation-Augmented Language Models (30 citations)
- Protein Representation Learning via Knowledge Enhanced Primary Structure Reasoning (17 citations)
- WikiWhy: Answering and Explaining Cause-and-Effect Questions (8 citations)
- STREET: A MULTI-TASK STRUCTURED REASONING AND EXPLANATION BENCHMARK (6 citations)
- Knowledge-in-Context: Towards Knowledgeable Semi-Parametric Language Models (2 citations)
Reasoning and chain of thought · 8 papers
- Self-Consistency Improves Chain of Thought Reasoning in Language Models (706 citations)
- Least-to-Most Prompting Enables Complex Reasoning in Large Language Models (317 citations)
- Automatic Chain of Thought Prompting in Large Language Models (233 citations)
- Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoning (108 citations)
- Complexity-Based Prompting for Multi-step Reasoning (73 citations)
- Language models are multilingual chain-of-thought reasoners (55 citations)
- Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thought (40 citations)
- Learning Math Reasoning from Self-Sampled Correct and Partially-Correct Solutions (2 citations)