Openi returned the GPT -4O update for Chat GPT, which causes the default personality of Chatboat to “excessively flattering or pleased -which is often called psychophantic” and that “psychophantic conversation can cause uncomfortable, annoying and anxious.” Says in a blog post.
According to the post, the company introduced a GPT -4O update last week, which included adjustments, aimed at improving the default personality of the model so that it could feel more intuitive and effective in many tasks. “Openi says he has begun to form the first model behavior with which he is described in Model Spec And teaches these models how to implement the principles in this speculation by adding a user signal to the thumb up / thumb on the answers of the chat GPT. “
But with a rolledback update, Open says, “We focused on short -term feedback, and did not fully calculate how users’ conversations with Chat GPT are ready over time.” This meant that “the GPT -4O quarreled in a response that was excessively helpful but invasive.”
The blog post says that the default personality of the Openi is designed to “reflect our mission and to respect, auxiliary, and respect various values and experiences,” says the blog post, but added that “not each of these desired features, such as trying to be useful or helpful.” The company says “a default cannot get every priority” for its 500 million weekly chat GPT users.
Openi will take further steps to re -describe the model’s behavior “, which” improving the basic training techniques and systems clearly indicates the model to remove the model from psychopulation “and indicates” extension methods “to give users feedback. The company says, “We also believe that consumers should gain more control over that chat is treated with GPT and, to the extent that it is safe and possible, if they do not agree with the default behavior, make adjustments.”
