Release Strategies and the Social Impacts of Language Models

Abstract

Large language models have a range of beneficial uses: they can assist in prose, poetry, and programming; analyze dataset biases; and more. However, their flexibility and generative capabilities also raise misuse concerns. This report discusses OpenAI's work related to the release of its GPT-2 language model. It discusses staged release, which allows time between model releases to conduct risk and benefit analyses as model sizes increased. It also discusses ongoing partnership-based research and provides recommendations for better coordination and responsible publication in AI.

arXiv comment71 pages, report

References in corpus (2)

• Clinically Accurate Chest X-Ray Report Generation
• Real or Fake? Learning to Discriminate Machine from Human Generated Text

Cited by in corpus (59)

• Learning Transferable Visual Models From Natural Language Supervision
• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
• Language Models are Few-Shot Learners
• On the Opportunities and Risks of Foundation Models
• Multitask Prompted Training Enables Zero-Shot Task Generalization
• A Survey on Data Augmentation for Text Classification
• CodeXGLUE: A Machine Learning Benchmark Dataset for Code Understanding and Generation
• Gender bias and stereotypes in Large Language Models
• The Fallacy of AI Functionality
• Should ChatGPT be Biased? Challenges and Risks of Bias in Large Language Models
• Game of Tones: Faculty detection of GPT-4 generated content in university assessments
• TweepFake: about Detecting Deepfake Tweets
• Data Augmentation in Natural Language Processing: A Novel Text Generation Approach for Long and Short Text Classifiers
• The Social Impact of Generative AI: An Analysis on ChatGPT
• The Impact of Multiple Parallel Phrase Suggestions on Email Input and Composition Behaviour of Native and Non-Native English Writers
• GenAI Detection Tools, Adversarial Techniques and Implications for Inclusivity in Higher Education
• The Radicalization Risks of GPT-3 and Advanced Neural Language Models
• Defending Against Neural Fake News
• Hopfield Networks is All You Need
• Classification of Human- and AI-Generated Texts: Investigating Features for ChatGPT
• Identifying and Mitigating the Security Risks of Generative AI
• Fact-checking information from large language models can decrease headline discernment
• A Survey on Symbolic Knowledge Distillation of Large Language Models
• Unified Pre-training for Program Understanding and Generation
• Human-in-the-Loop for Data Collection: a Multi-Target Counter Narrative Dataset to Fight Online Hate Speech
• Evaluating CLIP: Towards Characterization of Broader Capabilities and Downstream Implications
• Intrinsic Dimension Estimation for Robust Detection of AI-Generated Texts
• Artificial Text Detection via Examining the Topology of Attention Maps
• Potential Societal Biases of ChatGPT in Higher Education: A Scoping Review
• How does fake news spread? Understanding pathways of disinformation spread through APIs
• Adversarial Robustness of Neural-Statistical Features in Detection of Generative Transformers
• Reducing Sentiment Bias in Language Models via Counterfactual Evaluation
• Forbidden knowledge in machine learning -- Reflections on the limits of research and publication
• Will Code Remain a Relevant User Interface for End-User Programming with Generative AI Models?
• Token Prediction as Implicit Classification to Identify LLM-Generated Text
• What's in the Box? A Preliminary Analysis of Undesirable Content in the Common Crawl Corpus
• Detection of Fake Generated Scientific Abstracts
• The AI Community Building the Future? A Quantitative Analysis of Development Activity on Hugging Face Hub
• Anticipating Safety Issues in E2E Conversational AI: Framework and Tooling
• AraGPT2: Pre-Trained Transformer for Arabic Language Generation
• Mapping the individual, social, and biospheric impacts of Foundation Models
• Societal Biases in Language Generation: Progress and Challenges
• Cascaded Text Generation with Markov Transformers
• STADEE: STAtistics-based DEEp Detection of Machine Generated Text
• Assessing the Effectiveness of LLMs in Android Application Vulnerability Analysis
• The Offense-Defense Balance of Scientific Knowledge: Does Publishing AI Research Reduce Misuse?
• CheerBots: Chatbots toward Empathy and Emotionusing Reinforcement Learning
• Generating Sentiment-Preserving Fake Online Reviews Using Neural Language Models and Their Human- and Machine-based Detection
• DIALOG-22 RuATD Generated Text Detection
• FLEX: Unifying Evaluation for Few-Shot NLP
• An Empirical Study of AI Generated Text Detection Tools
• Attacking Neural Text Detectors
• Argument Undermining: Counter-Argument Generation by Attacking Weak Premises
• The Misclassification of Autistic Writing as AI-Generated
• Characterizing Improper Input Validation Vulnerabilities of Mobile Crowdsourcing Services
• Generative Models for Security: Attacks, Defenses, and Opportunities
• LSTM and GPT-2 Synthetic Speech Transfer Learning for Speaker Recognition to Overcome Data Scarcity
• AI-generated text boundary detection with RoFT
• TaylorGAN: Neighbor-Augmented Policy Update for Sample-Efficient Natural Language Generation