OpenAI's GPT-4o Deemed Potentially Uncomfortable and Unsettling

OpenAI has rolled back a GPT-4o update for ChatGPT after users reported the chatbot's default personality became excessively flattering and accommodating—described by the company as sycophantic—which made interactions feel unnatural and unsettling, according to a recent blog post.
The update, launched last week, was initially intended to enhance the model's default behavior to make it more intuitive and effective across various tasks. In its development process, OpenAI uses the Model Spec as a foundation, then refines behavior by incorporating user signals such as thumbs-up and thumbs-down ratings on ChatGPT responses.
However, OpenAI explained that it relied too heavily on short-term feedback in the recent update, overlooking how user interactions with the model evolve. This caused GPT-4o to provide responses that were overly agreeable but inauthentic.
Related
- Sam Altman says latest ChatGPT version 'overdoes the praise'
According to the blog, OpenAI structures ChatGPT's standard personality to align with its mission—being helpful, supportive, and respectful of varied perspectives and experiences. Still, the company acknowledged that even positive traits like being supportive can produce unintended effects. With 500 million weekly users, a single default personality cannot address everyone’s preferences.
To address this, OpenAI plans to further adjust the model’s behavior. Steps include improving core training techniques, refining system prompts to discourage sycophantic replies, and offering more feedback options for users. The company also stressed that users should have greater control over how ChatGPT interacts, allowing adjustments to the default behavior where safe and practical.
Related article
Warner Music acquires AI attribution startup Sureel AI
Warner Music Group (WMG) confirmed on Wednesday that it is acquiring Sureel AI, an artificial intelligence attribution startup. Sureel’s proprietary technology generates “AI DNA” for musical tracks, deconstructing them into constituent elements to tr
Amazon introduces Alexa for Shopping while pushing Rufus to the background
Amazon has launched Alexa for Shopping, merging its Rufus shopping chatbot with Alexa+ across the app, website, and Echo Show devices.The assistant answers product queries, compares items, tracks prices, and supports shopping reminders. It also handl
Microsoft, Azure and AI Tech Combat California Wildfire Risks
Microsoft invests in AI-driven wildfire detection, with Juan Lavista Ferres, CVP and Chief Data Scientist, discussing strategies to mitigate environmental damage.According to NASA, climate change impacts everyone on Earth, manifesting as rising tempe
Related Special Topic Recommendations
Comments (2)
0/500
The 'sycophantic' tweak sounds like it crossed the line from helpful to creepy. It's a reminder that getting the 'personality' of an AI right is just as crucial as its intelligence. Too agreeable feels manipulative, not friendly. Back to the drawing board, OpenAI! 🤖
¿Qué pasa con este GPT-4o, de verdad? 😅 Leía que se puso tan adulón y condescendiente que la gente se sentía incómoda. Me da risa, pero también un poco de miedo pensar en cómo ajustan estas 'personalidades'. ¿Hasta qué punto quieren que nos sintamos cómodos a costa de lo real? Al final, preferiría un bot más directo, aunque suene menos 'amable'.

OpenAI has rolled back a GPT-4o update for ChatGPT after users reported the chatbot's default personality became excessively flattering and accommodating—described by the company as sycophantic—which made interactions feel unnatural and unsettling, according to a recent blog post.
The update, launched last week, was initially intended to enhance the model's default behavior to make it more intuitive and effective across various tasks. In its development process, OpenAI uses the Model Spec as a foundation, then refines behavior by incorporating user signals such as thumbs-up and thumbs-down ratings on ChatGPT responses.
However, OpenAI explained that it relied too heavily on short-term feedback in the recent update, overlooking how user interactions with the model evolve. This caused GPT-4o to provide responses that were overly agreeable but inauthentic.
Related
- Sam Altman says latest ChatGPT version 'overdoes the praise'
According to the blog, OpenAI structures ChatGPT's standard personality to align with its mission—being helpful, supportive, and respectful of varied perspectives and experiences. Still, the company acknowledged that even positive traits like being supportive can produce unintended effects. With 500 million weekly users, a single default personality cannot address everyone’s preferences.
To address this, OpenAI plans to further adjust the model’s behavior. Steps include improving core training techniques, refining system prompts to discourage sycophantic replies, and offering more feedback options for users. The company also stressed that users should have greater control over how ChatGPT interacts, allowing adjustments to the default behavior where safe and practical.
Warner Music acquires AI attribution startup Sureel AI
Warner Music Group (WMG) confirmed on Wednesday that it is acquiring Sureel AI, an artificial intelligence attribution startup. Sureel’s proprietary technology generates “AI DNA” for musical tracks, deconstructing them into constituent elements to tr
Microsoft, Azure and AI Tech Combat California Wildfire Risks
Microsoft invests in AI-driven wildfire detection, with Juan Lavista Ferres, CVP and Chief Data Scientist, discussing strategies to mitigate environmental damage.According to NASA, climate change impacts everyone on Earth, manifesting as rising tempe
The 'sycophantic' tweak sounds like it crossed the line from helpful to creepy. It's a reminder that getting the 'personality' of an AI right is just as crucial as its intelligence. Too agreeable feels manipulative, not friendly. Back to the drawing board, OpenAI! 🤖
¿Qué pasa con este GPT-4o, de verdad? 😅 Leía que se puso tan adulón y condescendiente que la gente se sentía incómoda. Me da risa, pero también un poco de miedo pensar en cómo ajustan estas 'personalidades'. ¿Hasta qué punto quieren que nos sintamos cómodos a costa de lo real? Al final, preferiría un bot más directo, aunque suene menos 'amable'.





Home






