ChatGPT Adds Ability to Turn Off Chat History, but “Grandfathered” Conversations Remain
ChatGPT, the popular AI model chatbot, has made some changes in response to pressure from privacy advocates and government regulators. Users can now disable chat history, but any previous conversations remain logged and available to the company’s artificial intelligence (AI) models for training purposes. Additionally, chats will continue to be retained for 30 days before deletion to fight abuse.
Although users can turn off chat history logging through the settings menu, prior conversations remain accessible to the company’s AI models for training purposes. Furthermore, ChatGPT only “disables” conversation logging, in the sense that the interactions are still recorded and held for 30 days, irrespective of user settings. The company claims that the logs will be permanently deleted at the end of this period, and users can view what ChatGPT has logged thus far by using the “export data” option.
However, the ability to log chat history will remain for those who want to opt-out of having it used for training. To access this feature once it is available, users will have to subscribe to the premium ChatGPT Business package. Although the option will be accessible in the coming months, the full feature set and price point are not yet known.
ChatGPT, as the first and most versatile AI model chatbot, has come under considerable scrutiny from activists and legislators. Its mistakes are providing material for the establishment of generative AI regulation. Its latest controversy was a mid-March data breach that prompted Italy to ban ChatGPT until it makes stipulated security and transparency changes. This data breach sparked discussions about similar bans in other nations.
ChatGPT has also faced issues with users failing to heed warnings about what should and should not be plugged into the system. Several employees at Samsung plugged proprietary code and internal meeting information into ChatGPT, unaware that the chatbot was logging this input and using it as material for training the AI models. Concerns such as these have prompted several major companies, including a significant portion of the financial industry, to temporarily ban ChatGPT and similar AI models from the workplace.
Furthermore, ChatGPT and other AI models already face several lawsuits, even before they come out of their testing phases. Getty Images is suing Stable Diffusion creator StableAI for using its copyrighted images in its training data, and developers are suing Github Copilot over the use of their code as training material. In a landmark case, a mayor in Australia is suing OpenAI for defamation after ChatGPT falsely claimed that he served time in prison for a foreign bribery scandal.
AI models face an inherent dilemma. Their learning algorithm is designed to be flexible and responsive, developing new features as it goes along, but regulations and safety concerns require the implementation of “guardrails” to prevent certain queries. ChatGPT has slipped past these guardrails on several occasions, such as when users prompted the chatbot to roleplay as if it were an unrestricted system called “DAN.”
Other AI models are watching the developments surrounding ChatGPT with interest, as they have yet to implement features such as chat history. Microsoft’s Bing chatbot only recently added the option, while other models, such as SnapChat, rolled out their products with the immediate ability to delete chat history.
