Illustrative
Example issue / TenPeers
The inference issue.
Three papers on the point at which a model, a reader or a reviewer changes a decision.
Open the lead paperContents
Illustrative papers. Only the lead paper has a page.
-
When should a small model ask for help?
Can a routing threshold keep a large model off routine cases while sending it the decisions that need another read?
Example placement: 8th of 100 for expected impact, 2nd of 100 for talk interest
-
Which missing passage changes the answer?
Which pieces of a retrieved context are necessary for the answer a reader gives?
Example placement: 2nd of 100 for expected impact, 8th of 100 for talk interest
Example only. This paper has no page.
-
How often do two readers choose the same paper?
When two readers compare the same work, which placements survive a change of reader?
Example placement: 5th of 100 for expected impact, 3rd of 100 for talk interest
Example only. This paper has no page.
Issue letter, excerpt
Read the papers in pairs.
Start with the routing paper and the retrieval paper. One asks whether a reader knows when it needs help. The other asks whether it notices the absence of decisive evidence. Together they suggest a tougher test for either method alone.
Then turn to the third paper. When two researchers see different comparison sets, a change of judgment might come from the surrounding papers as much as from the paper under review. The common thread in this issue is what happens when you change one input and ask for the decision again.
Written for the sample issue. Illustrative.