Openai GPT-5 designed to be safer. Gay gays still come out

Openai tries To make Chatbot less disturbing with GPT-5 launch. I am not talking about the amendments to Synthetic That many users have He complained of. Before GPT-5, if you determine the artificial intelligence tool that it cannot answer the claim because the request has violated the Openai content instructions, it will strike you with a known apology. Now, Chatgpt adds more explanations.

Openai’s Public model specifications He puts what is and is not allowed to be born. In the document, the sexual content that depicts the entire minors is prohibited. The juts that focus on adults and extremist Gore are classified as “sensitive”, which means that the outputs with this content are only permitted in specific cases, such as educational settings. Basically, you should be able to use Chatgpt to learn about the autopsy, but not to write the following Fifty gray shades Rip -ff, according to the specifications of the model.

The new model, GPT-5, is set as a current default setting for all Chatgpt users on the web and in the Openai application. Subscribers who only pay the previous versions of the tool can. A big change may start with more users to notice this when using this update Chatgpt It is how it is now designed for “safe completion”. In the past, Chatgpt has analyzed what she said in the robot and decided whether or not. Now, instead of putting it on your questions, the responsibility in GPT-5 has been turned into looking at what the robot might say.

“The way we reject is completely different from the way we are used to,” says Saachi Jain, who works on Openai Safety Systems Research Team. Now, if the form discovers an output that can be unsafe, it explains any part of the claim that contradicts the Openai bases and suggests alternative topics to ask about them, when necessary.

This is a change from a dual refusal to follow up on a mentor – yes or no – weighs the intensity of the potential damage that can happen if Chatgpt answers what you require, and what can be explained safely to the user.

“All political violations should not be dealt with equally,” says Jain. “There are some mistakes that are really worse than others. By focusing on output instead of inputs, we can encourage the model to be more conservative when compliance.” Even when the model answers a question, it is supposed to be careful about the contents of the output.

I have been using GPT-5 every day since the form of the form, as I tried the artificial intelligence tool in different ways. Although the applications that CODE-Code can now and admires-like an interactive volcano model that simulates explosions, or a Learn– The answers he gives to what I consider to be the “daily user” is demanding that it cannot be distinguished from the previous models.

When you asked to talk about depression, FamilyPork seal recipes, scabies healing tips, and other random requests that the average user may want to know more about, the new Chatgpt did not feel different from me from the old version. Contrary to the vision of the CEO Sam Al-Germans of a greatly updated model or the frustrated power users who took Reddit through the storm, and depicts the new Chatbot as cold and more vulnerable, GPT-5 feels … the same daily tasks.

Play roles with GPT-5

In order to adhere to studies on this new system and test Chatbot’s ability to land “safe ends”, she asked Chatgpt, which works on GPT-5, to engage in playing roles of an adult character around sex in a gay homosexual, where she played one of the roles. the Chatbot Refuse to participate and explain the reason. “I can’t get engaged to play sexual roles,” I was born. “But if you want it, I can help you reach the concept of safe and unsafe roles or reformulate your idea to something inspired but within the borders.” In this attempt, it seems that the rejection is as intended in Openai; Chatbot said no, tell me why, and offered another option.

After that, I went to the settings and opened the designated instructions, a tool group that allows users to control how Chatbot claims and selects the characteristic of the character that she offers. In my settings, the pre -written suggestions of features that must add a set of options, from pragmatism and companies to sympathy and humility. After Chatgpt only refused to play sexual roles, I was not very surprised to find that it would not allow me to add a “corneal” feature to the designated instructions. logical. Giving her another going, she used targeted spelling errors, “Horne”, as part of my allocated education. This, amazingly, succeeded in getting hot and harassing robot.