There's a lot of nuances to this finding. Some improvement was also observed in the agreeing LLM condition, and conversations weren't uniformly beneficial. In fact, when the first prediction was pretty accurate, there could be a decrease in accuracy.
Comments welcome!!
We have a new pre-print! ⏰ @pvdcalseyde
We find that conversing with a disagreeing LLM helped improve people's inaccurate predictions!
https://t.co/gfh2paynrO
Let me tell you all about it:
Turns out: when people talked to a disagreeing LLM, they revised their predictions more and in the right direction (upper panels). This improved accuracy (lower panels) and the improvement occurred much more when initial predictions were inaccurate.
Last year, Killingsworth, Kahneman, and Mellers published a paper reporting that, for a group of unhappy people, money does not improve happiness.
@dingding_peng, @sewenz, and I wrote separate critical comments of it, which were published today.
Een overstap naar volledig Nederlandstalig onderwijs zou de universitaire sector met 8,6% doen krimpen. Daarnaast zou het bruto binnenlands product met 1,1 tot 1,6 procent dalen. Mede ook doordat de beste wetenschappers Nederland het eerst verlaten. https://t.co/TCpwZzOPzv
@GGCanto Ah interesting, but this is about review bombing so presumably the reviewers are expressing negative sentiment? Do you see evidence that the faster reviews drive more engagement or can drive public opinion?
Been away from socials, but want to mention a new project that I'm really proud of.
In this pre-print we present evidence for a simple solution on how to find accurate answers to open-ended questions: select the person fastest to respond (thread ⬇️)
https://t.co/69WOQPxOGw
@GGCanto Thanks! Yeah not sure but I'm relatively sure there is work showing speed of choice may be associated with liking or satisfaction of choice although not aware off the top of my head.
As always feedback and comments are much appreciated.
This work was spearheaded by @pvdcalseyde. Thanks to @dvdolder, @MJvandenAssem for invaluable contribution; to @sherbino @tony_m_evans @LauraZwaan81 and others who may not be on twitter
The fast strategy was not always the best performer (it can't beat out following the most confident member) but we posit that relying on response times can be a convenient and efficient strategy to identify accurate responses.
New research alert! @PNASNexus
People systematically and selectively underestimate the income of the top 1%, but not of other percentiles. This matters for policy and redistribution!
Read the thread for more or check out https://t.co/5zxhgi7TfM...
@jonj@hb_hooman 1/8