I think you’ll find this is very easy to do and very difficult to detect. (Think of an LLM populated with AI generated data.)
https://www.schneier.com/blog/archives/2023/05/on-the-poisoning-of-llms.html
On the Poisoning of LLMs
Interesting essay on the poisoning of LLMs—ChatGPT in particular:
Given that we’ve known about model poisoning for years, and given the strong incentives the black-hat SEO crowd has to manipulate results, it’s entirely possible that bad actors have been poisoning ChatGPT for months. We don’t know because OpenAI doesn’t talk about their processes, how they validate the prompts they use for training, how they vet their training data set, or how they fine-tune ChatGPT. Their secrecy means we don’t know if ChatGPT has been safely managed.
They’ll also have to update their training data set at some point. They can’t leave their models stuck in 2021 forever.
Once they do update it, we only have their word—pinky-swear promises—that they’ve done a good enough job of filtering out keyword manipulations and other training data attacks, something that the AI researcher El Mahdi El Mhamdi posited is mathematically impossible in a paper he worked on while he was at Google.
Tools & Techniques.
https://www.bespacific.com/what-is-chatgpt-and-why-does-it-matter/
What is ChatGPT and why does it matter?
ZDNET – Here’s what you need to know ….Updated: This AI chatbot’s advanced conversational capabilities have generated quite the buzz. We answer your questions.
How to use ChatGPT to:
(Related)
https://www.bespacific.com/how-to-be-on-the-lookout-for-misinformation-when-using-generative-ai/
How to be on the lookout for misinformation when using generative AI
Fast Company: “Until very recently, if you wanted to know more about a controversial scientific topic—stem cell research, the safety of nuclear energy, climate change—you probably did a Google search. Presented with multiple sources, you chose what to read, selecting which sites or authorities to trust. Now you have another option: You can pose your question to ChatGPT or another generative artificial intelligence platform and quickly receive a succinct response in paragraph form. ChatGPT does not search the internet the way Google does. Instead, it generates responses to queries by predicting likely word combinations from a massive amalgam of available online information. Although it has the potential to enhance productivity, generative AI has been shown to have some major faults. It can produce misinformation. It can create hallucinations —a benign term for making things up. And it doesn’t always solve reasoning problems accurately. For example, when asked if both a car and a tank can fit through a doorway, it failed to consider both width and height. Nevertheless, it is already being used to produce articles and website content you may have encountered, or as a tool in the writing process. Yet you are unlikely to know if what you’re reading was created by AI. As the authors of Science Denial: Why It Happens and What to Do About It, we are concerned about how generative AI may blur the boundaries between truth and fiction for those seeking authoritative scientific information. Every media consumer needs to be more vigilant than ever in verifying scientific accuracy in what they read. Here’s how you can stay on your toes in this new information landscape…”
No comments:
Post a Comment