OpenAI plans to improve ChatGPT's behavior to reduce biases, allow more user customization, and get more public input into decision-making about system defaults
We're clarifying how ChatGPT's behavior is shaped and our plans for improving that behavior, allowing more user customization …
OpenAI
Context & Ripple Effects
OpenAI’s plan framed ChatGPT behavior as both a product setting and a governance question: defaults would be revisited alongside user-level control and public input. Later, the company expanded custom instructions to all users, turning persistent user preferences into a broadly available feature.
That customization path continued with controls for ChatGPT’s presentation style in a later characteristics update and a memory summary page that lets users review and steer retained information. The arc is from setting a single assistant default toward making behavior more legible and adjustable.
First-order effects
OpenAI places ChatGPT’s bias-reduction work, system defaults, and user customization on the same product agenda, making its default behavior an explicit subject of design and public consultation.
ChatGPT users are positioned to receive more control over how the assistant responds rather than relying solely on OpenAI’s universal defaults.
Second-order effects
The later rollout of custom instructions makes those preferences persistent, shifting customization from one-off prompting into an account-level layer of the ChatGPT experience.
As OpenAI exposes more controls, it must distinguish user-selected behavior from the system defaults it continues to govern and explain.
Third-order effects
The subsequent combination of style controls and steerable memory points to assistants being managed through layered preferences—defaults, user instructions, and remembered context—rather than a single fixed persona.
If that pattern holds, AI-assistant governance will increasingly concern not only what a model permits, but which behavioral choices belong to providers, users, and public feedback processes.
The trend: Conversational AI is moving from provider-set assistant behavior toward layered user control paired with more explicit governance of system defaults.
How ChatGPT's behaviors are determined, and how we think it should work in the future — including thoughts on giving users much more control & early ideas around public input: https://openai.com/...
New update from @OpenAI 🔥 This paragraph gives you an idea of OpenAI's future direction: “We believe that AI should be a useful tool for individual people, and thus customizable by each user up to limits defined by society” https://openai.com/... https://twitter.com/...
this is going to take continual iteration—and lots and lots of societal input—to get right. to find the right balance, we will likely overcorrect several times, and find new edges in the technology. we appreciate the patience and good faith as we get to a better place!
ChatGPT latest blog (@openAI) proposes an improvement to remove bias on political and controversial topics. The introduction of a “public input” into the training loop. https://openai.com/... https://twitter.com/...
@sama Given the challenges you've outlined—"glaring and subtle biases," “define your AI's values”—you might want to expand your recruitment beyond software engineers. Nothing wrong with software engineers, but experts from other fields can bring particular insight on these issues…
One of the main reasons ChatGPT was released was to get feedback and stress test our systems. 🚧 Obviously, the first version is not perfect. If you have seen examples or are curious about AI bias, please read this post, very helpful and informative. https://openai.com/...
“[W]e are sharing a portion of our guidelines that pertain to political and controversial topics. Our guidelines are explicit that reviewers should not favor any political group. Biases that nevertheless may emerge from the process described above are bugs” #ethics #AI #ChatGPT h…
our current thoughts on hard questions about how AI systems should behave: 1) less biased defaults, 2) lots of user customization within very broad bounds, 3) public input on bounds and defaults https://openai.com/...
The researchers also propose stricter control over model access, including closing security vulnerabilities and restricting access to future models. https://www.vice.com/...
I'm sorry, is OpenAI training its models on internet complaints about women drivers? Quite an example to offer so un-ironically. https://openai.com/... https://twitter.com/...
leaked business plan from @OpenAI: 1 - create world's largest bullshit generator 2 - advocate for a clampdown on AI to restrict spread of toxic bullshit 3 - pat self on back for ensuring that AI benefits humanity https://www.vice.com/...