arXiv research summaries for Computation and Language from October 25, 2023. You can find summaries and links to each article here.
Today's Theme (LLM-Generated)
- Evaluating capabilities and limitations of large language models (LLMs) like ChatGPT and GPT-4 through tasks such as reasoning, summarization, and parsing.
- Improving alignment and control of LLMs through techniques like reinforcement learning from human feedback, supervised iterative learning, and contextualized prompt optimization.
- Enhancing fairness, interpretability, and transparency of models through bias mitigation, rationale extraction, and pretraining data detection.
- Advancing multilingual and multimodal models through pretraining, benchmarking, and adapting models like mBERT to new languages and modalities.
- Quantization and efficiency improvements for deploying large models through methods like post-training 4-bit floating point quantization.
