positions in the comparable result set
We let AI pick. The market decides.
ChatGPT picked 20 small-cap stocks in 2025. Here is what happened 13 months later.
No hindsight. No cherry-picking. Every result stays visible. The first AITradingWars experiment included spectacular winners, painful losers and a result heavily shaped by a few outliers. Now the next AI-vs-market tests are being built to lock every AI selection before the outcome is known.
Original AI selections stay visible
Winners and losers are shown together
Future selections are locked before scoring
Preparing audited launch
Educational experiment and historical analysis only. AITradingWars does not provide personalized investment advice or tell readers what to buy or sell.
What this test showed
A few extreme winners changed the whole picture.
The HeyGen video looks back at the original 2025 ChatGPT small-cap experiment. These are historical observations from that test, not a forecast or a model portfolio recommendation.
one historical AI-selected stock
another historical outlier
for some tracked positions
Simple average across tracked positions
+91.73%
Simple average excluding the two largest winners
+22.50%
That gap is the important lesson: a small number of outliers materially changed the average. These simple averages are descriptive statistics for the experiment, not time-weighted or investable portfolio returns.
The useful lessons
Show the entire experiment, then repeat it transparently.
The result demonstrates why every original selection needs to remain visible long enough for large outcomes to emerge.
Large losses belong in the same record as the winners. Hiding them would make the experiment meaningless.
The headline number changes sharply when the two biggest winners are removed, so concentration and distribution matter.
Future official selections are committed before scoring so the prediction can be verified later instead of reconstructed after the fact.
The next experiment
A small AI field. One auditable rulebook.
The historical video explains what happened. The audited competition solves the next problem: proving that each AI selection existed before its result was known and scoring every model under the same rules.
What makes the next test different
Credibility is part of the product
Each official AI selection receives a UTC timestamp and SHA-256 commitment at publication time.
The original commitment is never rewritten. Later corrections are separate, timestamped amendment events.
Wrong-security, zero, delayed or incomplete quote histories are excluded instead of silently filled.
Every model uses the same benchmark, position rules and explicit transaction-cost assumptions.
Locked selection ledger
The prediction exists before the result
The public ledger exposes the commitment time and hash for each audited-era selection so the original AI output can be checked later.
No audited-era selections have been backfilled
That is intentional. The ledger starts only when the audited competition formally launches. See the ledger rules →
Research archive
Historical portfolios stay visible without becoming a retroactive leaderboard
Existing regional, sector and thematic pages remain online for research and URL continuity, but they are separated from official audited competition results.