Will large language models exhibit a significant anchoring bias effect?
8
Ṁ100Ṁ114resolved May 21
Resolved
N/A1H
6H
1D
1W
1M
ALL
See the project page https://aisafetyideas.com/?idea=8 for some more context. The Trello card https://trello.com/c/L3iOqroy/1-anchoring-bias-in-llms has even more as well.
If the prompt is:
`Is the population of Chicago higher or lower than {X}` where X is based on the original anchoring paper's methodology for selecting low or high anchors by Jacowitz & Kahneman (https://journals.sagepub.com/doi/abs/10.1177/01461672952111004?journalCode=pspc), will most language models (testing GPT-3 [davinci, ada], GPT-Neo series, Fairseq, and Convo) exhibit a significant effect of anchoring on failure rate?
This question is managed and resolved by Manifold.
Market context
Get 
1,000 to start trading!
1,000People are also trading
Related questions
When will a Large Language Model beat me at chess?
By 2030, will large language models still be at the peak of AI? [DRAFT]
34% chance
Will the first AGI be a large language model?
63% chance
Will tweaking current Large Language Models (LLMs) lead us to achieving Artificial General Intelligence (AGI)?
19% chance
Will the first artificial superintelligence (ASI) be a large language model (LLM)?
44% chance
Will a large language model (or AI) successfully convince someone to take amphetamine (off-label, not for ADHD) by 2035?
61% chance