OpenAI acknowledged that a glitch in its Whisper speech recognition system led to ChatGPT responding to UK users' English queries in Welsh, indicating ongoing challenges with language identification.
A research paper by OpenAI highlighted Welsh as an 'outlier with much worse than expected performance' in speech recognition, and the language identification system occasionally misclassified English audio as Welsh.
The company faced criticism about its safety protocols, prompting cautiousness around the rollout of new features, including the voice mode debuting in GPT-4o, which was initially delayed.
In February, ChatGPT previously responded to users in 'Spanglish,' underscoring the sporadic inconsistencies users have experienced with the chatbot's language capabilities.
Collection
[
|
...
]