All About AI proposes an ongoing prediction challenge between GPT-5.6 and Opus 5. The first round asks each model to estimate Seoul's lowest temperature and the number of dissenting votes at a Federal Reserve meeting, then places small positions on the selected outcomes so later videos can score the forecasts.
The benchmarkA benchmark is a defined test or collection of tasks used to compare a system's performance under specified conditions. tries to prevent the models from copying the market consensusA prediction market lets participants trade contracts whose value depends on the outcome of a future event.. It rewrites each question to hide its Polymarket origin, removes Polymarket and Kalshi from search results, shows every possible outcome without prices and requires a probability distributionConfidence calibration measures whether an AI system's stated certainty matches how often its predictions or assumptions are actually correct. instead of a single yes-or-no answer.
Both models receive the same structured search interface and run one after the other in a cleaned repository. This keeps tool access comparable and reduces the chance that the second model will see the first model's output. One event per question also prevents several related market ranges from being counted as independent testsAn evaluation set is a collection of examples kept for measuring an AI system rather than training it..
Opus 5 selected the market-favored 26-degree temperature range and two dissenting votes. GPT-5.6 chose 25 degrees and zero dissenters, producing more divergent positions for the first round. The video does not yet establish which model is more accurate because the underlying events had not resolved when it was recorded.
The creator treats this as an early version of the scaffolding and notes that Opus matching the leading indicators could mean the price-obfuscation method still leaks cluesData leakage occurs when information unavailable or inappropriate at the intended decision point enters a model's training, evaluation or prediction process.. Future rounds are intended to improve the controls, test more models and accumulate enough outcomes for meaningful comparison. The SerpApi sponsor segment is omitted from this summary.
Watch on YouTube




