AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Pitfalls in Evaluating Language Model Forecasters

Daniel Paleka, Shashwat Goel, Jonas Geiping, Florian Tramèr

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 27 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
restrict questions to those resolved by the evaluation datebacktesting benchmarks · Forecasting benchmark with Jun 2023-Jan 2024 resolution windowshare of leaked early-resolved questions (%): at least 3.8%Hurt
restrict questions to those resolved by August 2024backtesting benchmarks · Forecasting benchmark resolved by Aug 2024share of trivial questions (%): at least 10%Hurt
generate questions retroactively from newsforecasting benchmarks · Binary news-generated questionsweak classifier accuracy (%): over 80%Hurt
reproduce MCQ pipeline with newer model and improved promptsMCQ forecasting dataset · Four-choice MCQclassifier accuracy (%): 80 -> 55Mixed
use date-restricted web searchretrieval-based forecaster · Backtesting retrievaltemporal leakageHurt
prompt GPT-4o with a later assumed knowledge cutoffGPT-4o · Knowledge cutoff testknowledge beyond cutoffHurt
give the LLM access to human market forecastsLLM vs human crowd comparison · ForecastBenchfairness of comparisonHurt
optimize for chance of winning instead of honest probabilitiesforecaster · Shared-latent-variable questionsexpected log score: -0.69N -> -0.64NHurt
report average Brier across mixed base ratesforecasting benchmark · Brier exampletotal Brier score: 0.125 vs 0.1Hurt
clamp probabilities under label noiselog-scored forecaster · Uniform true probability toy modelexpected log loss: 1/2 + eta -> 1/2 - eta + log(1+2 eta)Improved
filter to questions with at least 12 forecastersManifold benchmark · Manifold resolved questionsquestion availability: over 50% under 12 forecastersHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy