Reply to thread

Message: <blockquote data-quote="IndrajithGamage" data-source="post: 30259544" data-attributes="member: 581741">Well the improvement of systems doesn't seem to happen. Clearly this has been a known issue with their systems, long enough to get videos created on the topic, and multiple (easily thousands or even millions) of similar prompts being inserted, and at least a significant percentage of users reporting that the answer is wrong. Maybe there's a more sophisticated solution, but there needs to be some checkback for accuracy right? For example, if you ask about some known figure who has just died, ChatGPT basically answers from a database of sorts, which hasn't yet been updated with that information, so wouldn't point that out. However, you can force it to go online and read recent news articles and pick it up, almost always by asking "Are you sure?". In this case, instead of an online search, maybe such a call to reaffirm prompt after an obvious counting error like above could fall back to something like I said. If the easier way is to break up the word so that they get assigned to different tokens and if that provides accurate answers, can you explain to me why the immediate decision in the above example to search online and read most recent news articles can't be adapted to break down the word with spaces? <img class="smilie smilie--emoji" loading="lazy" alt="🤔" title="Thinking face    :thinking:" src="https://cdn.jsdelivr.net/joypixels/assets/6.6/png/unicode/64/1f914.png" data-shortname=":thinking:" /> This is not a hard coding as far as I can understand. It's just a method to reaffirm the accuracy of an answer, which at least in some aspects have already been built in.</blockquote>

Verification: Haya warak paha keeyada? (haya wadi kireema paha)

Top Bottom