OpenAI Shuts Down Their AI Detector Project
Previously OpenAI has stated that the ablity to tell AI text from human text will be important in the future. But they are more in the generation business than the detection business, and after continued “inaccuracies” in the detectors ability to separate AI and human text, they shut down their classifier project last week. That leaves a number of free and paid services, the current leader of which is Originality.ai, as the standard bearers for AI and plagiarism detection.
New Version of GPT Mastery Out Soon
Whic brings us to the next item of news: A new version of GPT Mastery! is imminent. Changes in this version include advice for automated blog writing, and advice for avoiding AI detectors. Existing customers will receive an updated copy.
Researchers Discover a Method for Generating Jailbreaks
Jailbreaks are techniques used to get around the safeguards that prevent AI models like ChatGPT from generating dangerous or harmful content (or often, merely “inappropriate” content). A team of researchers at Carnegie Mellon has discovered a way to use open source LLM models like LLaMA to generate strings that, when appended to a prompt, bypass the safeguards and allow the AI to discuss anything. The research paper describing the method and code can be found here. I would recommend exploring this only for the very curious or advanced user. Do not use jailbreaks on your primary OpenAI account, as it is a breach of their terms of service and could result in your account being suspended.
Cheers,
D.D.
GPT Mastery Team