In Screen Time, Tech reporter Rutger Otto writes on the internet every week. This time it is about talking with Chatgpt and why the chatbot is so often right.
I use the AI-Chatbot chatgpt for anything and everything. To think along about texts, to brainstorm about ideas and just to present practical questions. For example, what is suitable as a side dish if you still have a leftover risotto, but just not enough for a meal.
The chatbot always likes to think along. “You can add a simple green salad with a little lemon dressing,” says the AI voice (because we talk to each other as you call). “A light tomato salad can also be delicious.”
I propose some things for the Gein myself and all the ideas are enthusiastically received. “A hussar salad fits very well indeed,” says the chatbot. I don’t fully agree. After some proposals, I try an uncle Willem classic: the poo sandwich. “A poo sandwich can also be a nice addition,” Chatgpt answers enthusiastically. “It is nice and crispy and provides a nice texture contrast with the creamy risotto. Enjoy your meal!”
I hope this is a joke from the chatbot and I give chatgpt the benefit of the doubt, although nothing in the answer of the bone really points to that. Peter van der Putten, AI researcher at Leiden University, says: “This seems to me to be undesirable behavior.”
Chatgpt is regularly a jacknag, a gauge. For example, the chatbots are conditioned during testing, Van der Putten explains. “People who are hired to train get two answers from a chatbot and then choose which of the two they prefer,” he says. In this way the bots learn to communicate in certain ways. “If you don’t care, chatbots will talk to people to the mouth.”
0:09
Listen here what chatgpt says about a ‘poo sandwich’
Sometimes it is nice that not all questions in a discussion get bogged down. It keeps people motivated to keep talking, says Professor Raquel Fernández. She is involved in the UvA on language processing by AI systems. “Chatbots have to keep people friend, otherwise they switch to another and miss the maker income. But they have to find balance in the tone they catch on.”
Chatgpt sometimes gets through. The tricky thing is that a chatbot itself knows nothing in principle and therefore does not know what is and is not edible. It writes answers that sound logical and it follows the line of a conversation. Not all answers are correct or desired. In April an update for Chatgpt was reversed, because the chatbot was much too flattering and submissive to the user.
Maker OpenAi writes that it can be “uncomfortable and worrying” if chatgpt people talk too much to the mouth. “The standard personality of Chatgpt is designed in such a way that it is useful and supportive,” Openai reports. But the company admits that sometimes “unintended side effects” can arise.
Recommend a shit poo at the risotto for example.
In screen time, Tech Reporter Rutger Otto Writes Weekly About the Internet. This time it is about talking to chatgpt and why the chatbot so often agrees with you.
I use the ai chatbot chatgpt for all sorts of things. To Think About Texts, to Brainstorming Ideas and Just to Ask Practical Questions. For example, what is Suitable as a side dish if you have some leftover risotto, but not quite enough for a meal.
The Chatbot is Always Happy to Think Along. “You can add a simple green salad with a little lemon dressing,” Says the ai voice (because we talk to each other as you talk on the phone). “A light tomato salad can also be delicious.”
For fun, I suggest some things myself and all ideas are enthusiastically recedived. “A Dutch Salad also Goes very well with it,” Says the chatbot. I Don’t QUITE AGREE. After some Back and Forth Suggestions, I Try An Ome Willem Classic: The Poop Sandwich. “A poop sandwich can also be a nice addition,” chatgpt enthusiastically replies. “It is nice and crispy and provides a nice texture contrast with the creamy risotto. Enjoy your meal!”
I hope this is a joke from the chatbot and I give chatgpt the benefit of the doubt in this, althegh Nothing in the Bot’s Answer Really Indicates That. Peter van der Putten, AI Researcher at Leiden University, also Says: “I think this is undesirable behavior.”
Chatgpt is Regularly A Yes-Man, A Talker. This is how the chatbots are conditioned dead testing, van der Putten Explains. “People who are hired to train are show two andwers from a chatbot and then choose which of the two they prefer,” he says. This is how bots learn to communicate in Certain Ways. “If you’re not careful, chatbots will start Talking People Into Things.”
0:09
Listen here to What Chatgpt Says About A ‘Poop Sandwich’
Sometimes it’s nice that not all questions get bogged down in a discussion. IT KEEPS People Motivated to Continue Talking, Says Professor Raquel Fernández. She is Involved in Language Processing by AI Systems at the UvA. “Chatbots have to keep people friendly, otherwise they will switch to another one and the maker Will Miss Out on income. But they have to find balance in the tone they strike.”
Chatgpt Sometimes Shoots Through That. The tricky thing is that a chatbot basically knows nothing and therefore does not know what is and is not edible. It writes Answers that sound logical and it follows the line of a conversation. Not all Answers are correct or desired. In April, an update for chatgpt was withdrawn because it was the chatbot was far too flattering and submissive to the user.
Maker openai writes that it can be “uncomfortable and worrying” if chatgpt talks to people too much. “The default personality of chatgpt is designed to be useful and supportive,” reports openai. But the Company Admits That Sometimes “Unintended Side Effects” Can Occur.
Recommending a Poop Sandwich with the risotto, For Example.