Resolution Criteria

Predictions will be gathered from GPT-4 and GPT-4 with search in the week succeeding the tournament on all non-meta binary questions. Once the questions are all scored, this question resolves positively if the log score for GPT-4 with search is better than the log score for GPT-4 without search and negatively otherwise. If no official GPT-4 with search exists when we try to get the predictions, we will wait until up to November 17 for that functionality to return, and otherwise resolve this question as ambiguous.

Background Info

OpenAI's GPT-4 has a mode where it has access to internet search. By default it cannot access the internet or any tools.

To get the log score of a set of forecasts, take the sum of the negative logs of the probabilities given to the correct answers. Lower log scores are better.

This question was part of the OPTIC Fall 2023 Bay Area forecasting tournament (organized by @toms, @saulmunn, and @JingyiWang). I loved the questions, so I'm copying them (with permission) to Manifold (from private Metaculus). If resolution is unclear, I will copy whatever decision is made by OPTIC!