Artificial intelligence now beats some of the best human forecasters

Commenters debated whether AI outperforming top human forecasters is surprising or consequential. Some, like seanhunter and bagels, argued it's unsurprising given machine learning's long history as a powerful estimator and that non-LLM models have been used for forecasting for decades. Others warned the result may be fragile: johnecheck, qsbuilder and jacknews pointed to reflexivity - predictions changing the system they predict - and ddp26 and jesse_dot_id suggested AI could be especially weak at anticipating disruptive, unprecedented events. Several commenters (throwaway5752, croes) emphasized that human forecasters aided by AI might outperform either alone, while ratelimitsteve and adleyjulian raised luck and scoring issues in forecasting contests that could inflate apparent superiority.

Discussion also covered practical and ethical implications for markets and industry. glimshe proposed training LLMs to predict how mainstream AIs will invest, provoking counterpoints about meta-models (graypegg) and whether firms would quietly use top models to trade (rlt). Practitioners like gyanchawdhary, arn3n and wpasc noted barriers beyond algorithms - execution, data access, colocation, dark pools and overfitting - that limit monetization. Others questioned societal benefit, calling heavily algorithmic markets a game (superxpro12, in_absentia), and pointed to peripheral effects such as LLMs influencing baby names (varenc) or journalism styles; a few made light of prizes and fame.