Open says he will make changes About how he updates AI models that Power Chatgpt, following an incident that caused the platform to become too slanderous for many users.
Last weekend, after Openai put on a tweaked GPT-4O-the default model that supplies Chatgpt-social media users informed that the chatgpt began to respond in an excessively valid and enjoyable way. Quickly became a mimic. Users posted chats’ screenshots by welcoming all sorts of troubled, dangerous decisions and ideas.
In a post on X last Sunday, CEO Sam Altman recognized The problem and said that Openai would work on the “ASAP” corrections. On Tuesday, Altman announced The GPT-4O update was redesigned and that Openai was working on “additional corrections” on the model’s personality.
The company published a posthumous Tuesday and in a blog post on Friday, Openai extended to specific adjustments it plans to make in the model development process.
Openai says it plans to introduce an “Alpha phase” Opt-in for some models that would allow some Chatgpt users to try the models and give feedback before starting. The company also states that it will include explanations of “known restrictions” for future incremental updates in Chatgpt models and adjust the security revision process to officially examine “model behavior issues” such as personality, deception, reliability and illusion (that is, when a model).
“Going forward, we will contact preventively about the updates we make in the models in Chatgpt, whether” thin “or not,” Openai wrote in the blog position. “Even if these issues are not perfectly quantified today, we are committed to blocking the launches based on attorney measurements or quality signals, even when measurements such as the A/B test are good.”
The committed corrections come as more people turn to Chatgpt for tips. According to a recent survey With the legal funding of Financier Express, 60% of US adults used Chatgpt to look for tips or information. The increasing dependence on chatgpt – and the huge platform users base – increases the stakes when issues such as the extreme sycophancy arise, so as not to mention illusions and other technical weaknesses.
TechCrunch event
Berkeley, ca
|
June 5
Book now
As a moderate step, earlier this week, Openai said he would experiment with ways to let users give “real -time feedback” to “directly influence their interactions” with Chatgpt. The company also said it would improve techniques to direct models away from Sycophance, possibly allow people to choose from multiple models in Chatgpt, create additional security protective messages and expand evaluations to help detect issues beyond.
“One of the greatest lessons fully recognizes the way people have started using chatgpt for deep personal tips – something we didn’t see so long a year ago,” Openai continued in her blog. “At that time, this was not a primary focus, but as AI and society is co-opted. It has become clear that we must deal with this use with great care.
