For years now, anyone who has used a computer has been agreeing to terms and conditions without really understanding the rights and limitations they are agreeing to. Even if the software or service clearly states that you must read the license, not many people do because they’re long, tedious, and quite difficult for most people to understand.
With the introduction of generative AI, however, these licenses are becoming easy for normal people to understand. Not only can you get a summary of your main rights and obligations, but you can also craft specific questions and get an answer based on the actual text.
At first, with the initial introduction of ChatGPT, this wasn’t feasible because of character limits on your inputs. More recently, however, Claude 2 and Poe.com launched with the ability to attach PDF or TXT documents allowing the AI to look over comprehensive licenses with lots of text.
The restriction that you can only share your account with people in your household was there before the new fee came in but few people knew about it because people skip over the terms.
You may have bought a Netflix subscription to keep your media consumption above board and then unwittingly have broken the rules by sharing the account with friends or family in another house, putting you back in a bit of a precarious position.
Another very important aspect of licenses is your rights. For example, there is an AI image maker online called Dezgo with has a complex license document attached outlining the rights and restrictions.
If you have a business and you want to know if you can use one of the images for commercial purposes, it would normally require your to get through a quagmire of legal jargon. Even if you do find the specific clause, the technical language could still make your understanding of your rights a bit ambiguous.
With generative AI tools that support file uploading, however, you can simply ask it ‘Does the license allow me to use the images for commercial purposes?’ and using the contents of the document, provides an accurate answer.
Using generative AI to assess concrete information in this way is actually a more effective way to use it. One of the issues it has with open-ended queries, which most people pass to generative AI now, is that it can fill in gaps in its knowledge with hallucinations – made-up bits of information.
While there are efforts to tackle hallucination through a method called Chain of Thought (CoT), that work is still ongoing and is not foolproof. With a concrete set of data to work on, it can get through it without having to make things up and provide better answers.
Overall, the usefulness of generative AI being able to assess these big documents will vary from person to person. Some will go on flagrantly violating the terms of streaming services while others, and businesses, will be able to ensure they’re not breaking any costly commercial restrictions.
Have you come across any useful applications of generative AI that people don’t seem to be talking about all that much? Let us know in the comments.