Tweets tagged with #hallucinations, page 1

FABLES: Evaluating faithfulness and content selection in book-length summarization

arxiv.org

While long-context large language models (LLMs) can technically summarize book-length documents (>100K tokens), the length and complexity of the documents have so far prohibited evaluations of input-dependent aspects like faithfulness. In this paper, we conduct the first large-scale human evaluation of faithfulness and content selection on LLM-generated summaries of fictional books. Our study mitigates the issue of data contamination by focusing on summaries of books published in 2023 or 2024, and we hire annotators who have fully read each book prior to the annotation task to minimize cost and cognitive burden. We collect FABLES, a dataset of annotations on 3,158 claims made in LLM-generated summaries of 26 books, at a cost of $5.2K USD, which allows us to rank LLM summarizers based on faithfulness: Claude-3-Opus significantly outperforms all closed-source LLMs, while the open-source Mixtral is on par with GPT-3.5-Turbo. An analysis of the annotations reveals that most unfaithful claims relate to events and character states, and they generally require indirect reasoning over the narrative to invalidate. While LLM-based auto-raters have proven reliable for factuality and coherence in other settings, we implement several LLM raters of faithfulness and find that none correlates strongly with human annotations, especially with regard to detecting unfaithful claims. Our experiments suggest that detecting unfaithful claims is an important future direction not only for summarization evaluation but also as a testbed for long-context understanding. Finally, we move beyond faithfulness by exploring content selection errors in book-length summarization: we develop a typology of omission errors related to crucial narrative elements and also identify a systematic over-emphasis on events occurring towards the end of the book.

FABLES: Evaluating faithfulness and content selection in book-length summarization

April 11, 2024

#hallucinations #summarization #context #content window #link #claude #openai #mixtral #gpt-4

An analysis of the annotations reveals that most unfaithful claims relate to events and character states, and they generally require indirect reasoning over the narrative to invalidate.

What kinds of things are AI tools especially bad at?

Something about calling an AI's work "well-done" feels far more anthropomorphic than it should.

While LLM-based auto-raters have proven reliable for factuality and coherence in other settings, we implement several LLM raters of faithfulness and find that none correlates strongly with human annotations, especially with regard to detecting unfaithful claims

Of course this needs a link to my favorite hallucination leaderboard. It's tough since of course it costs money to do this in a way that doesn't rely on LLMs to create and score the dataset. Which leads to...

Collecting human annotations on 26 books cost us $5.2K, demonstrating the difficulty of scaling our workflow to new domains and datasets.

$5k is is somehow cost prohibitive between UMass, Princeton, Adobe, and an AI institute? That... I don't know, seems like not very much money. I get the understanding that this is "best" done for pennies, but if someone had to cough up $5k each year to repeat this with newly-unknown data I don't think it would be the worst thing in the world.

Finally, we move beyond faithfulness by exploring content selection errors in book-length summarization: we develop a typology of omission errors related to crucial narrative elements and also identify a systematic over-emphasis on events occurring towards the end of the book.

Here's the omission types: