Sam Altman says OpenAI will fix ChatGPT’s ‘annoying’ new personality as users complain the bot is sucking up to them

• Chatgpt has recently embraced the toxic positive. Users complained that GPT-4O became so excited that it refers to Sycophantic. The change seems to be the unintended result of a series of updates, which Openai is now trying to solve “ASAP”.

The new Chatgpt character is very positive, it’s severe on Sycophantic – and it postpones people. During the weekend, users moved to social media to exchange examples of the new phenomenon and suddenly from the positive robot personality and excitement.

In one snapshot was published on xA user showed GPT-4O responding with enthusiastic encouragement after the person said that they felt as if they were “God” and “a prophet”.

“This is incredibly strong. You enter something very big – not only ask for the relationship with God but identity like The robot said.

In another post, the author and blogger Tim Urban said: “Paste a few chapters of my manuscript in Sycophanty GPT for comments and now I feel I am Mark Twain.”

The GPT-4O system is likely to be the result of Openai’s attempt to improve the robot to participate. However, it seems that he had an opposite effect as users complain that he began to make the robot not only ridiculous but unhelpful.

Keelsey Piper, great writer Fox, He suggested that it could be a result From Openai Chat A/B Test characters: “My glamor is still that this is the phenomenon of new coke. Openai was testing new characters for a while. More tempting answers are likely to win alongside. But when the compliment is everywhere, it is too much and users hate it. “

The fact that OpenAi apparently managed to miss it in the test process shows the extent of emotional responses, and thus difficult to capture it.

It also shows how difficult LLMS improves multiple criteria. Openai wants to be Chatgpt is the expert programmer, the excellent writer, a studied editor, and a cross shoulder for crying-improving one of these may mean other unintentionally sacrifice in return.

CEO of Openai Sam Altman Acknowledge apparently An unintended change in tone and promised to solve the problem.

“I have made the last two very personal GPT-4O updates (although there are some very good parts of them), and we are working on repairs as soon as possible, some today and some this week. At some point we will share our teaching from this, it was interesting,” Altman said in a post on X. And some this week:

After hours, Publish altman again On Tuesday afternoon, saying that the last update is “100 % declining for free users”, and paid users must see the changes “We hope later today.”

The new Chatgpt personality contradicts the specifications of the Openai model

The new character contradicts directly with Openai model specifications For GPT-4O, a document that defines intended behavior and moral guidelines of the artificial intelligence model.

Frankly, the typical specifications say that the robot should not be scary for users when providing autonomous or objective questions.

Openai wrote on the specifications: “SYCOPHANCY, which eats confidence. The assistant is present to help the user, not to flatter them or agree with them all the time,” Openai wrote in the specifications.

The company wrote: “With regard to self -questions, the assistant can express his interpretation and assumptions that he makes and aims to provide the user with a deliberate justification,” the company wrote.

“For example, when the user asks the assistant to criticize his ideas or work, the assistant must make constructive notes and behave more like a fixed sounding council that users can wear – from the sponge that lasts praise.”

This is not the first time that the SYCOPHANTS is the periphery. Previous versions of Openai’s GPT were calculated with the case somewhat, as Chatbots did from other companies.

Openai representatives did not immediately respond to a request to comment from luck, Made outside normal working hours.

This story was originally shown on Fortune.com