OpenAI, the company behind the famous ChatGPT chatbot, has announced some major improvements to the underlying language model, GPT-3. The new features include behavior customization, multimodal capabilities, and enhanced safety and security.
With the rapid growing popularity of AI chatbots
since the release of ChatGPT
, there is increased pressure on OpenAI and other AI developers to their systems immune against abuse and manipulation. If this will be possible, remains to be seen.
Behavior customization for GPT-3
The announced behavior customization will allow users to fine-tune GPT-3's responses according to their preferences and needs. For example, users can adjust the tone, style, politeness, and humor of GPT-3's outputs. Some of this kind of finetuning can already be seen in the Compose feature of Microsoft´s new Edge browser, where users can choose the tone for AI generated text output.
According to OpenAI, users will also be able to specify the level of detail, complexity, and accuracy they want from GPT-3's answers.
Behavior customization is achieved by using a simple interface that lets users provide feedback to GPT-3 on how well it performed on a given task. Users can also provide examples of desired outputs or keywords that guide GPT-3's behavior. By learning from user feedback and examples, GPT-3 can adapt its behavior over time and generate more personalized and relevant responses.
Multimodal Capabilities, Enhanced Safety and Security
Multimodal capabilities enable GPT-3 to handle inputs and outputs that involve more than one modality, such as text, images, audio, or video. For instance, users can ask GPT-3 to generate captions for images, summarize videos, transcribe speech, or create visualizations from data.
New enhanced safety and security measures aim to ensure that GPT-3 is used responsibly and ethically by preventing harmful or malicious outputs. For example, users can filter out profanity, hate speech, personal information, or sensitive topics from GPT-3's outputs. Users will be able to monitor and audit GPT-3's activity and report any issues or concerns. To achieve this, OpenAi uses human reviews, automated detection, and user feedback.
Tip of the day: Tired of Windows´s default notification and other system sounds? In our tutorial we show you how to change windows sounds or turn off system sounds entirely.