4 ms·
ChatGPT and A.I. ‘hallucinations’ will ever go away: ‘This isn’t fixable’
- archo 3y agoExperts starting to doubt AI ‘hallucinations’ will go away: ‘This isn’t fixable’ HN: Discussion - (35-comments - 2023-08-02) : https://news.ycombinator.com/item?id=36977096 https://news.ycombinator.com/item?id=36977096
- 3cats-in-a-coat 3y agoIt’s too tiresome to argue with these people. Let reality show them what’s possible and not. Models already can tell apart falsehoods but they are too small right now to do so reliably. GPT4 has the world’s knowledge in its model and it has the capacity of a mouse’s brain. It’s vastly overstretched. So it’ll hallucinate.
- jqpabc123 3y agoThey’re really just sort of designed to predict the next word Is "intelligence" merely a matter of data collection and statistics? I think not. But a poor simulation of it might be. And this is where we are today. It is where AI has always been.
- 3cats-in-a-coat 3y agoYou are one of those people who will hold onto something intangible that makes them better than machines. But you could neither explain what it is nor prove it scientifically.
- jqpabc123 3y agoYou are one of those people who will hold onto some technology that makes machines better than people. But you could neither explain what it is nor prove it scientifically. My basic contention is this: It's not possible to engineer/build something that has yet to be defined. And "intelligence" has yet to be defined. But I'm pretty sure that if/when it is, the definition will involve something more than a statistical database or a word prediction engine.
- 3cats-in-a-coat 3y agoI don’t have to prove anything. People are being replaced as we speak.
- jqpabc123 3y agoPeople being replaced is nothing new.
- 3cats-in-a-coat 3y agoNo. But what’s new is we’re out of places to go. Intellectual work and art were our last hiding places.
- Cheyana 3y agoI asked GPT-4 what it knows about hallucinations. “ Yes, in the context of AI, “hallucination” refers to a phenomenon where a machine learning model generates or perceives something that isn’t present in the input data. This can occur in various tasks like image recognition, natural language processing, or generative tasks. Hallucinations in AI models may result from biases or artifacts in the training data, or from the model overfitting to noise or unimportant details. They can lead to incorrect or misleading outputs and are a subject of ongoing research and concern in the field of AI. Ensuring that models generalize well from the training data to unseen examples is a key challenge in preventing hallucinations.”