TechcraftingAI NLP

TechcraftingAI NLP

By Brad EdwardsTechnology
Download on the App Store

All TechcraftingAI NLP episodes

  • Ep. 51 - November 14, 2023

    arXiv research summaries for Computation and Language from November 14, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


    Today's research themes (AI summary)

    • Evaluating and improving the reasoning capabilities of large language models through prompts and benchmarks.
    • Enhancing multi-lingual language models, especially for low-resource languages.
    • Detecting and reducing hallucinations and factual inconsistencies in language model outputs.
    • Applying language models to specialized domains like medicine, law, speech, and search.
    • Developing more efficient and accelerated decoding methods for language model generation.
    • 1 hr 20 min
    • Ep. 50 - November 13, 2023

      arXiv research summaries for Computation and Language from November 13, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


      Today's research themes (AI summary)

      • Improving safety, robustness, and interpretability of large language models through techniques like red teaming, uncertainty estimation, and neuron analysis.
      • Evaluating capabilities and limitations of large language models on diverse tasks like logic puzzles, knowledge integration, syntax sensitivity, and cultural concept popularity ranking.
      • Enhancing text generation with constraints and control through approaches like progressive editing, inner monologue prompting, and verifiable retrieval.
      • Studying entrainment, bias, and memorization in models for speech, text classification, and summarization.
      • Developing resources and models for low-resource languages and domains like Greek, Tamil, Tagalog, and scientific information extraction.
      • 1 hr 18 min
      • Ep. 49 - November 12, 2023

        arXiv research summaries for Computation and Language from November 12, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


        Today's research themes (AI summary)

        • Enhancing reasoning capabilities of large language models through techniques like chain-of-thought prompting or multi-agent discussions.
        • Improving cross-lingual transferability of models through techniques like knowledge sharing, alignment, and calibration.
        • Evaluating alignment of large language models with human values through adversarial evaluation frameworks.
        • Incorporating knowledge graphs and ontologies to inject external knowledge and improve language model pretraining.
        • Leveraging speech information and other multimodal signals in addition to text to enhance language model capabilities.
        • 37 min
        • Ep. 48 - November 11, 2023

          arXiv research summaries for Computation and Language from November 11, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


          Today's research themes (AI summary)

          • Improving language model performance on out-of-distribution data through techniques like counterfactual data augmentation.
          • Mitigating language model bias and toxicity at inference time without impacting translation quality.
          • Using prompts and demonstrations to enable few-shot learning for tasks like event extraction and question answering.
          • Analyzing model robustness through frameworks to systematically test perturbations and model variations.
          • Leveraging cross-lingual retrieval to improve multilingual model zero-shot performance.
          • 23 min
          • Ep. 47 - November 10, 2023

            arXiv research summaries for Computation and Language from November 10, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


            Today's research themes (AI summary)

            • Fine-tuning and benchmarking large language models for specialized domains like medicine and finance.
            • Evaluating safety, alignment, and Memorization in large language models.
            • Incorporating knowledge into large language models.
            • Theoretical understanding of large language model capabilities.
            • Novel applications of large language models like scientific hypothesis generation.
            • 23 min
            • Ep. 46 - November 9, 2023

              arXiv research summaries for Computation and Language from November 9, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


              Today's research themes (AI summary)

              • Improving neural network models for tasks like machine translation, grammatical error correction, and text representation through techniques like knowledge distillation and quantization.
              • Using large language models for summarization, dialogue, and question answering, and analyzing their capabilities and limitations.
              • Enhancing fairness, safety, and human alignment of large language models through methods like counterfactual awareness and human evaluations.
              • Leveraging semantic information like frames and events across documents and sentences for tasks including source validation and argument extraction.
              • Developing end-to-end approaches for spoken language tasks by incorporating speech recognition capabilities.
              • 28 min
              • Ep. 45 - November 8, 2023

                arXiv research summaries for Computation and Language from November 8, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


                Today's research themes (AI summary)

                • Evaluating large language models on human language acquisition and developmental plausibility
                • Using large language models to study human memory and cognitive processes
                • Analyzing biases in large language models across different personas and demographic groups
                • Developing more efficient large language model architectures and training methods
                • Assessing robustness and contamination issues in benchmarks used to evaluate large language models
                • 34 min
                • Ep. 44 - November 7, 2023

                  arXiv research summaries for Computation and Language from November 7, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


                  Today's research themes (AI summary)

                  • Improving language model performance through prompt engineering and auxiliary memories
                  • Evaluating model capabilities and biases through targeted perturbations and human preference alignment
                  • Extracting structured knowledge from unstructured text through multi-task learning and graph representations
                  • Analyzing model architectures to identify shared computational patterns across tasks
                  • Mitigating model harms through adversarial attacks and vulnerability analyses
                  • 48 min
                  • Ep. 43 - November 6, 2023

                    arXiv research summaries for Computation and Language from November 6, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


                    Today's research themes (AI summary)

                    • Improving text-to-SQL and knowledge base question answering with large language models via in-context learning.
                    • Leveraging large language models for zero-shot bilingual app review mining and abstractive summarization.
                    • Detecting artificially generated text with efficient model ensembles.
                    • Incorporating categorical labels and syntactic information to enhance biomedical named entity recognition.
                    • Analyzing model architectures for modeling human annotation variation and subjectivity in tasks like argument quality assessment.
                    • 32 min
                    • Ep. 42 - November 5, 2023

                      arXiv research summaries for Computation and Language from November 5, 2023. You can find summaries and links to each article ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠here⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠.


                      Today's research themes (AI summary)

                      • Relation extraction methods to identify triples and reduce redundancy
                      • Topic modeling for social media and short text
                      • Datasets for fake news and misinformation detection
                      • Temporal sequencing of historical documents
                      • Morphological analysis and language modeling for low-resource languages
                      • 23 min