OpenAI's ChatGPT has come under scrutiny for a variety of issues as reported by users and experts. Google's DeepMind has raised concerns about the potential for ChatGPT to leak training data and violate user privacy, stating that it can 'violate privacy'. Engineers at OpenAI have attempted to address 'GPT laziness' by removing the pressure for short responses and adding an automatic continuation for messages that still have content to write. However, these changes have led to broken code blocks and reduced capacity, resulting in slower performance. Harmful biases in ChatGPT's training data have been observed in HR applications such as job posts, candidate outreach, and performance feedback. The premium version of ChatGPT, GPT-4, has been specifically criticized for being 'completely broken' for code outputs, prompting users to look for alternatives. Amidst these challenges, OpenAI has been criticized for incentivizing developers to create engaging chatbots, a model that has been considered harmful in the past 'via @opinion'. The company is also dealing with legal issues, having failed to defeat a libel suit related to fabrications by ChatGPT.
OpenAI must defend ChatGPT fabrications after failing to defeat libel suit https://t.co/8QEhqVF5Qv
OpenAI will pay developers based on how engaging they can make their chatbots. @parmy reminds us that model has proved harmful in the past (via @opinion) https://t.co/f4VYrSmkgv
It seems that OPENAI has been trying to fix the lazy Gpt and has created the broken gpt in ChatGPT. https://t.co/ZkIGd7x1NC
ChatGPT (Premium, GPT-4) still completely broken for code outputs. Not good. Looking for alternatives now. https://t.co/nPdiRlxlcS
ChatGPT's training data contains harmful biases that show up in the output. We've seen this across a variety of HR applications, including job posts, candidate outreach, and performance feedback. Learn how to use AI without destroying your DEI efforts: https://t.co/AWdNZazxya https://t.co/L44zXUcO1T
ChatGPT engineers are struggling OpenAI -โFixesโ GPT laziness by removing short response pressure, and adding a 2nd message auto continue if there is still content to write -breaks code blocks cuz the second message closed the block in the first -nukes capacity, slow as hell https://t.co/7EhB4pLcs2
ChatGPT can leak training data, violate privacy, says Google's DeepMind https://t.co/6Rsz69yVyX