5 ms·
If the text fits within the context window most models are pretty reliable at summarizing texts effectively.
by reidjs 2y ago
If the text fits within the context window most models are pretty reliable at summarizing texts effectively.
- nicklecompte 2y agoThey’re good at summarizing text convincingly if you didn’t read the original document, but GPT-4 still makes a ton of basic factual mistakes: https://arxiv.org/abs/2402.13249 https://arxiv.org/abs/2402.13249 I think it will be several decades - maybe longer - before you can throw an arbitrary 20-page document into a computer program and get an accurate one-page summary. People need to take seriously that transformer LLMs don’t understand human language.