Uri Katz
@urikauri.bsky.social
NLP PhD student @biunlp.bsky.social Working on IR, Search, Reasoning. Intern @ai2.bsky.social.
https://katzurik.github.io/
https://katzurik.github.io/
Since (Fisher) information is the inverse of variance, it also decreases.
November 22, 2024 at 12:34 PM
Since (Fisher) information is the inverse of variance, it also decreases.
When evaluating a model, we want the variance of the estimation to be low. Variance decreases as the sample size increases. if observations (questions) share correlated trajectories, the effective sample size decreases, leading to an increase in variance.
November 22, 2024 at 12:33 PM
When evaluating a model, we want the variance of the estimation to be low. Variance decreases as the sample size increases. if observations (questions) share correlated trajectories, the effective sample size decreases, leading to an increase in variance.
If a paragraph includes a finite number of signals (facts, arguments, or properties), then the number of questions based on the chain of those signals is limited and will likely include parallel reasoning trajectories, I guess.
November 22, 2024 at 9:56 AM
If a paragraph includes a finite number of signals (facts, arguments, or properties), then the number of questions based on the chain of those signals is limited and will likely include parallel reasoning trajectories, I guess.
Congratulations, Luca! Could you share a bit about the lessons learned that were surprising for the team?
November 22, 2024 at 9:35 AM
Congratulations, Luca! Could you share a bit about the lessons learned that were surprising for the team?
Did it answered back?
November 20, 2024 at 7:14 AM
Did it answered back?