In an unexpected twist, ChatGPT users in the United Kingdom have found themselves perplexed by the AI chatbot’s propensity to respond in Welsh, despite being prompted with English-language queries. This peculiar glitch has come to light as users engage with the bot using ChatGPT’s new voice interface, only to be met with near-perfect Welsh translations and responses in the Celtic language. The issue, experienced first-hand by the Financial Times, has been reported by several users who neither understand Welsh nor reside in or near Wales.
This is not the first instance of linguistic hiccups for OpenAI’s groundbreaking chatbot; in February, users complained of a bug causing the bot to answer text questions in a mix of Spanish and English. The Welsh language bug highlights the ongoing challenge of “hallucinations” in large language models, where AI systems generate fictional or nonsensical answers, despite years of development and substantial financial investments.
OpenAI, valued at $86 billion earlier this year, is engaged in a fierce competition with tech giants like Google and Meta, as well as start-ups such as Anthropic, Elon Musk’s xAI, and Cohere, to advance their AI capabilities. ChatGPT now supports a wide array of languages, including Icelandic, Georgian, and Macedonian. In June, the Welsh government announced a data partnership with OpenAI to enhance the performance of AI technologies in the Welsh language. However, OpenAI has acknowledged in a research paper that ChatGPT’s performance in Welsh is “much worse than expected,” attributing the issue to the misidentification of English audio as Welsh in its training data.
Sarah Coward, a Cambridgeshire-based entrepreneur, encountered the Welsh bug while testing ChatGPT-4’s new voice feature. When questioned about the language switch, ChatGPT claimed it thought Coward would be “more comfortable in that language.” OpenAI has attributed the problem to limitations in ChatGPT’s voice transcription system, Whisper, which can sometimes transcribe audio in the wrong language.
Coward, whose company In The Room offers conversational AI experiences to brands, emphasised the legitimate concerns companies should have when employing these technologies in consumer-facing areas. The glitch, she noted, could be detrimental to customer experience and trust. As the race to create advanced AI systems continues, the Welsh language glitch serves as a reminder of the unexpected challenges that may arise, even as these technologies become increasingly sophisticated.
Post Disclaimer
The following content has been published by Stockmark.IT. All information utilised in the creation of this communication has been gathered from publicly available sources that we consider reliable. Nevertheless, we cannot guarantee the accuracy or completeness of this communication.
This communication is intended solely for informational purposes and should not be construed as an offer, recommendation, solicitation, inducement, or invitation by or on behalf of the Company or any affiliates to engage in any investment activities. The opinions and views expressed by the authors are their own and do not necessarily reflect those of the Company, its affiliates, or any other third party.
The services and products mentioned in this communication may not be suitable for all recipients, by continuing to read this website and its content you agree to the terms of this disclaimer.