Frontier AI models outperform human experts on earnings prediction for the first time

Automated excerpt

Surprise correlation measures if the surprise (i. e. actual − consensus) is correlated with the predicted surprise (prediction − consensus). Surprise correlation for AI models is increasing with more capable models, with the latest models significantly outperforming the bias-corrected consensus. Earnings prediction error for AI models is decreasing with more capable models, with all models outperforming consensus and the best model, GPT-6 Astra, outperforming even the bias-corrected consensus.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.