OpenAI has now made ChatGPT more human than it ever has been before

When OpenAI's ChatGPT was released in 2022, it was already jarring to see how much the chatbot could engage in fluent conversation. Three years later, the humanizing aspects only continue to evolve.

2

The latest update from OpenAI emphasizes personalization and customization. You can now tailor the chat box to the dimensions of tone (witty, supportive) and functional roles (lab assistant, companion). You'll find a settings window that provides a few options to contextualize the conversation:

Functionally, this allows me to hardwire the chatbot into the harsh, but honest tone of Arnold Schwarzenegger. Highlighting that it needs to call me 'Steve', and that I'm an office worker with a bonked knee. The perfect recipe for fitness advice!

Much of this functionality was present in previous iterations - you could prompt it to take on a certain voice (Arnold) or traits (harsh but honest), and the 'memory' feature would allow it to retain information (the bum leg). However, this update more so represents a step in making this function more consistent and easier to use. You'll no longer need to keep re-prompting it to achieve the output you want consistently.

That update also comes with a focus on transparency. In May of 2024, OpenAI released their 'model spec', which essentially exists to help you understand why and how ChatGPT responds the way it does. As we become more adept with manipulating LLM chatbots, understanding these underpinnings as we utilize the more advance features is becoming more and more important.