Staff and students are advised that the output from ChatGPT may be particularly odd or unusual for the next 24 hours as OpenAI try to roll back an upgrade performed late last night. A spokesperson from OpenAI explained that last night’s upgrade was based on an advanced set of training data which included James Joyce’s ‘Finnegans Wake’. However, it is now believed that Joyce’s writing has confused ChatGPT’s LLM (Large Language Model) as it was noticed early this morning that ChatGPT’s responses were increasingly containing non-standard punctuation, puns, portmanteau words, and had an almost dreamlike quality to them. A technical expert from OpenAI said that, “while ChatGPT has been trained on a number of classic novels, it simply wasn’t able to parse Joyce’s prose in Finnegans Wake, and this seems to have entirely undermined ChatGPT’s ability and willingness to write in a conventional style.” OpenAI have said that they are confident that the problem will be speedily resolved but have advised users to carefully check ChatGPT’s output for what they are referring to as ‘linguistic oddities and anomalies’ for the next few days.
