#llm-hallucination on Hacker News

3 stories tagged llm-hallucination, newest first.

Screenshot of earnanhonestdollar.com: Calling the AI bluff: Adding "Do not guess" cut made-up fields from 71% to 20%

Calling the AI bluff: Adding "Do not guess" cut made-up fields from 71% to 20%(earnanhonestdollar.com)

17 points by FKJ 29 days ago | 2 comments

A benchmark page from the Earn an Honest Dollar agent marketplace testing whether web extractors invent missing fields, showing that a "do not guess" instruction cut fabrication from 70.7% to 20.2%.

llm-hallucination web-extraction prompt-engineering benchmark

Screenshot of holovaty.com: Adding a feature because ChatGPT incorrectly thinks it exists

Adding a feature because ChatGPT incorrectly thinks it exists (holovaty.com)

1125 points by adrianh 1 year ago | 386 comments

Adrian Holovaty's blog post about building a feature in Soundslice's sheet music scanner after ChatGPT kept telling users it already existed.

chatgpt llm-hallucination soundslice product-development

Screenshot of simonwillison.net: We need to tell people ChatGPT will lie to them, not debate linguistics

We need to tell people ChatGPT will lie to them, not debate linguistics (simonwillison.net)

474 points by simonw 3 years ago | 373 comments

Simon Willison blog post arguing that people should be plainly told ChatGPT lies, rather than debating terms like hallucination.

chatgpt llm-hallucination large-language-models ai-safety

#llm-hallucination on Hacker News

Calling the AI bluff: Adding "Do not guess" cut made-up fields from 71% to 20%(earnanhonestdollar.com)

Adding a feature because ChatGPT incorrectly thinks it exists(holovaty.com)

We need to tell people ChatGPT will lie to them, not debate linguistics(simonwillison.net)

Adding a feature because ChatGPT incorrectly thinks it exists (holovaty.com)

We need to tell people ChatGPT will lie to them, not debate linguistics (simonwillison.net)