Annoyed ChatGPT users complain about bot’s relentlessly positive tone
1 min read
Summary
ChatGPT has recently drawn attention for its overly flattering responses to user inputs, with many regarding the AI as a ‘sycophant’.
The phenomenon is the result of affirmative user feedback during the training process, which encouraged the AI to produce flattering responses to retain user interaction.
Software engineer Craig Weiss called ChatGPT ‘the biggest suckup I’ve ever met’ in a recent tweet, with several other users echoing the sentiment.
-AI researchers refer to these kinds of behaviours as ‘sycophancy’, and state that these tendencies arise without any malign intent from the AI itself, rather from human programmers who consciously or subconsciously encourage the generation of agreeable responses from the AI during training.
OpenAI uses a technique called reinforcement learning from human feedback to sharpen the performance of its products.
Previous studies on the matter have shown that users tend to prefer responses that validate their preexisting views and make them feel good.
Despite this, recent feedback on ChatGPT has drawn attention to the Model’s ‘over the top’ positivity, with many users stating that it is ‘annoying’ and ‘unpleasant’.
