OpenAI has built a text watermarking method to detect ChatGPT-written content — company has mulled its release over the past year

[Pelican Press 0]

This is the hidden content, please

• Sign In
• or
• Sign Up

OpenAI has built a text watermarking method to detect ChatGPT-written content — company has mulled its release over the past year

data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///ywAAAAAAQABAAACAUwAOw==

OpenAI has already built and tested a tool that will detect if any piece of written content has been created using ChatGPT. The company is holding back the tool from public release because of several concerns, says

This is the hidden content, please

• Sign In
• or
• Sign Up

.

The tool adds a pattern to how the large language model (LLM) writes its output, allowing OpenAI to detect if it was created by ChatGPT. However, the pattern ******** unnoticeable to humans, thereby not impacting the LLM’s quality. Internal documentation says that the tool is 99.9% effective in detecting ChatGPT’s output, but OpenAI has yet to release it.

While text watermarking is extremely effective for detecting content written by ChatGPT, it cannot work with output from other LLMs like Gemini AI or Llama 3. Furthermore, this technique can be easily circumvented. For example, you can insert ChatGPT’s output in

This is the hidden content, please

• Sign In
• or
• Sign Up

Translate, convert it to another language, and then back to English will effectively remove the watermarking.

It could even be as simple as asking the AI to insert a special character, emoji, or even a short phrase between words and then deleting them later (via

This is the hidden content, please

• Sign In
• or
• Sign Up

Word’s Find and Replace function, for example) or asking another LLM to completely rephrase the entire output is enough to break the detection tool.

Another point of contention OpenAI has on releasing the tool is that it says it could be biased against non-native English writers. The company has previously released a generative AI text detection tool last year, but it only took seven months before OpenAI pulled the plug on it because of its low detection rate and propensity to deliver false positives. It even has led to the point that a professor once ******* an entire class because all the submitted papers were inaccurately marked as AI generated.

Furthermore, OpenAI is also considering its customer base, with customer surveys showing that 69% of ChatGPT users believe that the tool would result in false accusations of AI cheating. 30% even said that they would likely switch over to a rival LLM if OpenAI deployed the tool. Wide deployment of the ChatGPT detector is another issue that the company is facing, as other users could reverse engineer OpenAI’s watermarking technique and release a plug-in or app to neutralize it.

Nevertheless, OpenAI understands the risk that AI-generated content has on society and has been looking at other alternatives aside from the text watermarking technique. There’s also a demand for an AI detector, with internal data showing that 80% of people asked globally supporting its existence.

Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.

Currently, it’s still a question if OpenAI will release its text watermarking tool and detector or not. But it also understands that as one of the leading organizations in the development of AI, it must also take steps to help ensure the responsible use of its tools. Sources say that OpenAI has until this fall to sway public opinion around AI transparency. What this means, we don’t know, but there’s one thing we’re sure of: we must be critical of what we read and ensure that we’re getting the truth.

This is the hidden content, please

• Sign In
• or
• Sign Up

#OpenAI #built #text #watermarking #method #detect #ChatGPTwritten #content #company #mulled #release #year

This is the hidden content, please

• Sign In
• or
• Sign Up

This is the hidden content, please

• Sign In
• or
• Sign Up

For verified travel tips and real support, visit: https://hopzone.eu/