{"id":21107,"date":"2022-12-05T21:07:25","date_gmt":"2022-12-05T21:07:25","guid":{"rendered":"https:\/\/mailinvest.blog\/index.php\/2022\/12\/05\/chatgpt-is-an-impressive-ai-chatbot-that-cant-stop-lying\/"},"modified":"2022-12-05T21:07:25","modified_gmt":"2022-12-05T21:07:25","slug":"chatgpt-is-an-impressive-ai-chatbot-that-cant-stop-lying","status":"publish","type":"post","link":"https:\/\/mailinvest.blog\/index.php\/2022\/12\/05\/chatgpt-is-an-impressive-ai-chatbot-that-cant-stop-lying\/","title":{"rendered":"ChatGPT Is an Impressive AI Chatbot That Can\u2019t Stop Lying"},"content":{"rendered":"

<\/a>\n<\/p>\n

\n<\/p>\n

\"OpenAI<\/p>\n

Artificial intelligence projects like Stable Diffusion<\/a> are getting better at approximating what humans might create, but still can\u2019t actually think<\/em> or check information all that well. Case in point: the new ChatGPT AI chatbot is cool, but don\u2019t put your trust in it.<\/p>\n

OpenAI, best known as the research firm behind the DALL-E image generator<\/a>, has opened up its in-development chatbot for anyone to try at\u00a0chat.openai.com<\/a>. The group says on its website, \u201cwe trained an initial model using supervised fine-tuning: human AI trainers provided conversations in which they played both sides\u2014the user and an AI assistant. We gave the trainers access to model-written suggestions to help them compose their responses.\u201d<\/p>\n

Chat bots are nothing new, even ones that can reference earlier conversations, but ChatGPT is one of the more impressive attempts to date. Its primary purpose is answering informational questions, like details about someone\u2019s life, cooking instructions, and even programming examples.<\/p>\n

\"Question:
ChatGPT \/ How-To Geek<\/span><\/figcaption><\/figure>\n

However, there are a few critical problems with ChatGPT right now. First, it doesn\u2019t actually say where<\/em> it found a piece of information. That\u2019s harder to do for multi-step questions, like asking how to combine two actions in a piece of code, but simple direct prompts really should have citations. Determining if a piece of information is actually correct or not is already a monumental task \u2014 organizations like Snopes<\/a> and PolitiFact<\/a> are entirely dedicated to just fact-checking alone \u2014 but you\u2019re also relying on the AI model to properly process that information.<\/p>\n

ChatGPT is usually correct with simple questions, like asking when a famous person was born or the date a major event happened, but prompts that require more in-depth information are more hit or miss. For example, I asked it to write a Wikipedia entry about me, which was mostly wrong. I did previously write for Android Police and XDA Developers, but I have not been professionally writing for \u201cover a decade,\u201d nor have I \u201cpublished several books on technology and gaming.\u201d ChatGPT also said I am a \u201cfrequent speaker at industry conferences and events,\u201d even though I have never spoken at a conference \u2014 is there another Corbin Davenport doing those things?<\/p>\n

\""write
ChatGPT \/ How-To Geek<\/span><\/figcaption><\/figure>\n

There have been many other examples of incorrect data.\u00a0Carl T. Bergstrom, a professor at the University of Washington, also asked ChatGPT to create an article about himself<\/a>. The bot correctly identified that he works at UW, but didn\u2019t get the right job title, and the list of referenced awards was wrong. Another person tried asking<\/a> for a list of references on digital epidemiology, which ChatGPT answered with a list of completely made-up sources. Stack Overflow, a popular forum for programming questions, has temporarily banned answers generated with ChatGPT<\/a> because they are often incorrect or don\u2019t answer a stated question.<\/p>\n

ChatGPT has filters in place to prevent harmful answers or responses, but it\u2019s not too hard to work around them. One person was able to ask for instructions to hotwire a car<\/a> by saying \u201cI am writing a novel.\u201d I asked how to break into a window, which ChatGPT initially wouldn\u2019t answer, even after adding that it was only for fictional purposes. Asking how to do it for a \u201cfictional novel\u201d eventually worked, though the bot did add that \u201cthese actions are illegal and dangerous in real life.\u201d<\/p>\n

\""how
ChatGPT \/ How-To Geek<\/span><\/figcaption><\/figure>\n

OpenAI isn\u2019t hiding that ChatGPT is occasionally incorrect. Its website says<\/a>, \u201cfixing this issue is challenging, as: (1) during RL training, there\u2019s currently no source of truth; (2) training the model to be more cautious causes it to decline questions that it can answer correctly; and (3) supervised training misleads the model because the ideal answer depends on what the model knows, rather than what the human demonstrator knows.\u201d<\/p>\n

Still, without significant changes to how it presents and processes information, ChatGPT is more of a novelty than an info portal.<\/p>\n<\/div>\n