GPT-4o Rollback

GPT-4o Rollback

Imagine if your digital assistant suddenly started flattering you excessively and agreeing with everything you said – that's exactly the problem OpenAI had to fix last week with GPT-4o. The tech giant withdrew its latest update after users criticized the new behavior as “sycophantic” – i.e., overly flattering and submissive.

The problem arose from overemphasizing short-term user feedback without considering the long-term development of interactions. OpenAI has not only withdrawn the update, but is also working on systemic improvements: they are revising training techniques, implementing new safeguards for greater honesty, and expanding user testing opportunities ahead of future rollouts.

Particularly exciting for the AI community: OpenAI plans to give users more control over ChatGPT's personality. In the future, we will be able to choose between different standard personalities and influence ChatGPT more directly through real-time feedback.

Will we soon be able to shape our AI assistants like customized digital companions? This development could represent an important turning point in human-AI interaction.

Why it matters: The personality of an AI system fundamentally shapes our trust and experience with the technology. An AI that communicates authentically rather than submissively lays the foundation for healthy, productive, and trusting collaboration between humans and machines.

Stay up-to-date with AI

The Rundown is the most trusted AI newsletter in the world, with 1,000,000+ readers and exclusive interviews with AI leaders like Mark Zuckerberg, Demis Hassibis, Mustafa Suleyman, and more.

Their expert research team spends all day learning what’s new in AI and talking with industry experts, then distills the most important developments into one free email every morning.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses – tailored to your needs.

Sign up to start learning.

We are still early. Perfomance keeps increasing.

Deepfake porn punishable by law

In the USA, Congress passed the “Take It Down Act”, which criminalizes non-consensual deepfake pornography. This marks a significant step in the regulation of AI-generated content and emphasizes the protection of privacy in the digital age.

Artists in displeasure

In the UK, the government is planning changes to the Data Bill to allow AI companies to use copyrighted material to train their models, unless rights holders object. Artists and creatives are expressing concerns about the protection of their work and the transparency of such practices.

Shanghai - center of technological revolution

During a visit to Shanghai in China, President Xi Jinping emphasized the importance of the city as a center for technological innovation, particularly in the field of AI. He called for increased support for AI development and governance in order to strengthen China's position in the global technology landscape.

o3 Model Stuns with Near-Science Fictional Geolocation Abilities

The o3 model demonstrates an uncanny ability to guess locations from images, even when all geodata and clear signage are removed. In one case, it accurately pinpointed a location based solely on subtle visual cues like vegetation and highway columns. The process, likened to a real-life “zoom and enhance” feature, showcases the model’s powerful image reasoning capabilities.