GPT-5 is said to be more pleasant now

OpenAI updates GPT-5 for more warmth, addressing rollout issues to improve user experience.

: OpenAI has announced an update to its GPT-5 model, aiming to make it warmer and friendlier after a less-than-smooth launch. Many users initially preferred the older GPT-4o, prompting the company to address concerns with subtle changes. Tests show no increase in sycophancy compared to the current personality of GPT-5. The company aims to improve user interactions, with a focus on further developments beyond GPT-5.

OpenAI has recently rolled out an update to their newest model, GPT-5, aiming to make interactions more pleasant by altering its response style. The launch itself was admittedly rocky, as confirmed by CEO Sam Altman, who described the process as "a little more bumpy than we’d hoped for." User feedback highlighted a preference for the older GPT-4o model, which spurred OpenAI to make these adjustments in their latest version.

The updates are designed to make GPT-5 feel more approachable. This includes the addition of 'good question' or 'great start' acknowledgments, avoiding unnecessary flattery, which had been a conscious decision by OpenAI to keep the model authentic and engaging without becoming sycophantic. These changes, while subtle, are intended to enhance the overall user experience according to OpenAI's recent social media post.

During informal discussions at a dinner event with journalists, the focus was meant to be on OpenAI's future plans beyond GPT-5. However, as reported by Max Zeff, much of the attention remained on the challenges surrounding the model’s launch. VP Nick Turley remarked that the GPT-5 had a direct approach, which was intended to be softened with the new update, making it feel friendlier.

These optimized modifications follow internal tests which interestingly noted no increase in the model’s sycophancy levels when compared to its earlier versions. This reassures users that despite the introduction of warmth and friendliness, the integrity of responses remains central to GPT-5’s design.

OpenAI's ongoing efforts represent their commitment to refining AI models by directly addressing user concerns while focusing on authentic and effective communication.

Sources: TechCrunch, OpenAI

Generative AIinnovationMachine Learning