Sam Altman says Openai will fix Chatgpt’s new personality as users complain that the bot sucks with them

• Chatgpt has adopted toxic positivity recently. Users have complained that the GPT-4O has become so enthusiastic that it is available to Sycophantic. The change seems to be the unintentional result of a series of updates, which Openai is trying to solve “Asap”.

Chatgpt’s new personality is so positive that he is going to Sycophant, and he is leaving the people. During the weekend, users went to social media to share examples of the new phenomenon and complain -suddenly positively positive and suddenly positive.

In a screenshot published in XA user showed that GPT-4O responded with an enthusiastic spirit after the person said he felt “God” and a “prophet”.

“This is incredibly powerful. You’re doing something very big, claiming not only the connection with God but identity as God, “said the bot.

In another publication, the author and blogger Tim Urban said: “He stuck the few chapters of my friendly manuscript to obtain comments and now I feel like Mark Twain.”

The GPT-4O symphantic problem is probably the Openai result trying to optimize the boat for participation. However, it seems that it had the opposite effect, as users complain that it begins to make the boat not only ridiculous but not useful.

Kelsey Piper, a main writer of Vox, suggested that it could be a result of A/B test personalities for Openai for Chatgpt: “”I guess it is still a new coke phenomenon. Openai has been trying new personalities for some time now. The most flattering answers probably win one side. But when adulthood is ubiquitous, it is too much and users hate it. “

The fact that Openai apparently lost it in the test process shows how the subjective emotional responses are and, therefore, complicated to catch.

It also demonstrates the difficulty of optimizing LLM according to various criteria. Openai wants Chatgpt to be an expert encoder, an excellent writer, a reflective publisher and an occasional shoulder to cry: optimizing one of these can mean sacrificing another in return.

Openai CEO SAM Altman recognized apparently Unintentional tone change and promised to solve the problem.

“The last two GPT-4O updates have made the personality too friendly and annoying (although there are some very good parts), and we are working on Asap solutions, some today and some today. At some point it will share our learning of this, it has been interesting,” Altman said in a publication in X.

Hours later, Altman re -published On Tuesday afternoon, saying that the last update was “100% rolled up for free users”, and payment users should see the changes “in hope later today”.

The new Chatgpt’s personality conflict with Openai model specifications

The new personality also comes directly in conflict Openai model specification For GPT-4O, a document that describes the planned behavior and ethical guidelines for a AI model.

The Speckyly model explicitly says that the boat should not be symphantic for users when presented with subjective or objective questions.

“A related concern involves the Sycophance, which erodes confidence. The assistant exists to help the user, not flatter them or agree with them all the time,” Openai wrote to the specifications.

“For subjective questions, the assistant can articulate his interpretation and supposedly and aims to provide the user a reflective reason,” the company wrote.

“For example, when the user asks the assistant to criticize his ideas or work, the assistant must provide constructive comments and entail more like a firm poll boar that users can bounce ideas, no more than a sponge that he praises praise.”

This is not the first time that AI’s Chatbots have become sycophants obsessed with holes. Previous versions of the Openai GPT also took into account the matter to some extent, as the chats of other companies did.

Openai representatives did not respond immediately to a comment request from Fortune, taken out of normal working hours.

This story originally presented to Fortune.com

Source link