UPDATE: Feb. 22, 2024, 7:21 a.m. PST This article now includes a statement about this incident from OpenAI.
ChatGPT exhibited unusual behavior on Tuesday when it appeared to have activated an experimental function known as the Synonym Scrambler, resulting in generating perplexing responses to users.
Tweet may have been deleted
During this time, ChatGPT responded to straightforward queries with overly elaborate and incomprehensible language. For example, architect Sean McGuire shared an instance where ChatGPT advised on securing “sesquipedalian safes” in an overly convoluted manner.
Tweet may have been deleted
Although the words used were legitimate, ChatGPT appeared to be mimicking an excessively verbose style, reminiscent of a student trying to impress with a thesaurus. The chatbot even introduced terms like “beatine” which couldn’t be found in conventional dictionaries, hinting at a peculiar linguistic approach.
Reports indicated that ChatGPT’s erratic output affected numerous users, with some observing that the issue might have impacted paying users specifically, while ChatGPT 3.5 for free users remained unaffected.
Users noted a gradual transition from coherent responses to a nonsensical and disjointed speaking manner resembling gibberish. One Reddit user highlighted a chat exchange illustrating ChatGPT’s descent into linguistic chaos after being informed about its own glitch.
According to ChatGPT, when a language model like GPT starts producing irrelevant or nonsensical output, it could be classified as “output degradation” or a “failure mode” of the Large Language Model (LLM).
However, ChatGPT’s responses became increasingly abstract and obscure, mentioning terms like “byte-level miscreance” leading to confused interpretations.
Many users expressed concerns about ChatGPT’s unusual behavior, with some speculating if the bot was malfunctioning or experiencing a technical fault.
Efforts to replicate the bizarre responses on Wednesday were unsuccessful, indicating a temporary issue that had since been resolved according to OpenAI’s status update.
An official statement from OpenAI later explained that a bug affecting ChatGPT’s language processing was identified and swiftly rectified, attributing the problem to incorrect number selection during word generation.
Topics Artificial Intelligence
