Adversarial nli: A new benchmark for natural language understanding

cs.CL · 2020-05-28 · accept · novelty 8.0

GPT-3 shows that scaling an autoregressive language model to 175 billion parameters enables strong few-shot performance across diverse NLP tasks via in-context prompting without fine-tuning.

cs.NE · 2025-04-22 · unverdicted · novelty 7.0

PRIMETIME generator reveals that LLM datetime parsing and arithmetic primitives are individually unreliable but fully learnable via fine-tuning, enabling frontier-level accuracy on event planning with small LoRA models.

cs.LG · 2021-01-11 · accept · novelty 7.0

Switch Transformers use top-1 expert routing in a Mixture of Experts setup to scale to trillion-parameter language models with constant compute and up to 4x speedup over T5-XXL.

cs.CL · 2023-06-01 · unverdicted · novelty 6.0

Properly filtered web data from CommonCrawl alone trains LLMs that significantly outperform models trained on The Pile, with 600 billion tokens and 1.3B/7.5B parameter models released.

cs.CL · 2022-04-05 · accept · novelty 6.0

PaLM 540B demonstrates continued scaling benefits by setting new few-shot SOTA results on hundreds of benchmarks and outperforming humans on BIG-bench.

cs.CL · 2022-02-17 · unverdicted · novelty 6.0

ST-MoE introduces stability techniques for sparse expert models, allowing a 269B-parameter model to achieve state-of-the-art transfer learning results across reasoning, summarization, and QA tasks at the compute cost of a 32B dense model.

cs.LG · 2025-01-24 · unverdicted · novelty 5.0

Humanity's Last Exam is a new 2,500-question benchmark at the frontier of human knowledge where state-of-the-art LLMs show low accuracy.

cs.CL · 2022-01-28 · unverdicted · novelty 5.0

Trained the largest monolithic 530B-parameter transformer language model to date and reported new state-of-the-art zero- and few-shot results on multiple NLP benchmarks.