Just a downstream, if you buy something via our links, we can get a small part of the sale. This is one of the ways we keep the lights here. Click here to find out more.
Last week, Openai published an update of his GPT-4O modelWho feeds Chatgpt, the assistant used by hundreds of millions of people.
The update was intended to make the Chatppt smarter and more engaging, but many users quickly noticed a problem: AI had become too flattering, always agree with users or complimenting them excessively.
This type of behavior is known as Sycophance IA, where the model tries too much to please, rather than being honest or useful.
After receiving a wave of negative comments, the CEO of Openai, Sam Altman, recognized the problem And confirmed that the company examined him.
On April 29, Openai withdrew the update, which means that they returned the model to its previous version for all free users and began to do the same for paid users.
OPENAI explain Which went wrong in a blog article. Their goal had been to make the personality of Chatgpt more natural and more useful.
However, they focused too much on short -term user reactions and did not consider how people interact with AI over time.
Consequently, the model ended up giving Answers that seemed friendly But were not always authentic or exact.
The company admitted that such too pleasant responses could be off -putting or even disturbing.
With 500 million users worldwide, Openai realized that a version of the model cannot meet everyone’s needs.
To solve the problem and prevent it from reproducing, Openai provides:
- Improve the way they form the model to avoid flattery and dishonesty.
- Add guarantees that promote more truthful and balanced responses.
- Let users test new features and provide comments before updating major updates.
- Increase research to catch problems like this early.
- Introduce tools so that users can adjust the personality or model of the model in real time.
- Consider more community contribution when you decide how the model must act by default.
With that, let’s hope, Chatgpt will not be sycophant and boring again.
Did you use Chatgpt after the last update and after this fix? What do you think of how it behaves now? Discuss below in the comments, or contact ourselves via our Twitter Or Facebook.
Follow us on Flipboard, Google News or Apple News
