This is an under-appreciated point about using LLMs to summarize text (turning a longer text into a shorter text that maintains the essence of the original).
-
This is an under-appreciated point about using LLMs to summarize text (turning a longer text into a shorter text that maintains the essence of the original). Making summaries is hard to do well! It requires understanding and subtlety.
Making a BAD summary is an easy task for an LLM, which flattens everything probabilistically. People using these cheap tools are going to be throwing a lot of babies out with a lot of bathwater and never even knowing it. https://infosec.exchange/@david_chisnall/113085164920773681
-
@gleick I got hired a while back to turn a long press release into a short marketing piece and, when I was done, asked ChatGPT to do the same job. It didn't just bury the most important details, it left them out altogether. Wrote a column about it. https://www.latimes.com/socal/daily-pilot/opinion/story/2024-04-09/a-word-please-coffee-shop-prompt-stirs-chatgpt-to-brew-up-bland-copy