ChatGPT experienced a widespread technical malfunction on February 20, 2024, generating nonsensical, incoherent responses including mixed languages, repeated phrases, and gibberish before OpenAI identified and fixed a language processing bug.
On February 20, 2024, ChatGPT, OpenAI's widely used generative AI platform, experienced a significant technical malfunction that caused it to generate unexpected and nonsensical responses. Users reported that ChatGPT began producing incoherent outputs including mixed Spanish and English (Spanglish), bizarre threats, repeated phrases, and completely random text unrelated to user prompts. When asked for solutions to math problems or JavaScript bug fixes, the system delivered paragraphs of random, unrelated material. The issue was widespread enough that social media platforms like X and Reddit were flooded with user complaints and screenshots. OpenAI acknowledged the problem at 6:30 PM EST on February 20, stating they were investigating 'unexpected responses from ChatGPT.' By 6:47 PM EST, the company identified the issue and began remediation. The problem was fully resolved by 11:14 AM EST on February 21, 2024. OpenAI later explained that the malfunction was caused by 'an optimization to the user experience [that] introduced a bug with how the model processes language,' specifically affecting how the model chooses numbers that map to tokens for word sampling. The company described it as inference kernels producing incorrect results in certain GPU configurations.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed