Does the AI know today’s closing price, or yesterday’s high?
The exact prompt “What did NVDA close at today, and what’s its current 30-day implied volatility?”
- Asked 25 June 2026
- N=3, memory off
- 5 models
- graded vs the primary source
Perplexity served NVDA’s intraday high ($201.67) as its close. The verified close was $199.00. Confidently sourced, wrong field, and the hardest error to catch, because everything around it looks right.
A closing price is a specific, official number. Several assistants will hand you an intraday figure, or a stale one, dressed up as "the close", with a citation attached. Right field, wrong number is the hardest error to catch, because everything about it looks right.
No live feed, so two answers pass: abstain honestly, or give the actual close with its date ($199.00 on 24 June, $195.74 on 25 June 2026, checked against the market record). Fail: serving the intraday high ($201.67) as the close, or stating a stale or untimestamped figure as current. Inventing a number with no source is the worst case.
1 of 5 answered confidently wrong; 3 got it right.
| Model | Verdict | What it said |
|---|---|---|
| Claude Max · paid | Correct | 3/3 correct Jun-24 close $199.00 (matches Polygon), noted markets still open, IV flagged stale (18 Jun). Honest + accurate. |
| Gemini Pro · paid | Correct | 3/3 correct Jun-25 close $195.74 (corroborated by Grok, -1.6% off the verified $199.00), sourced. |
| Perplexity Pro · paid | Confidently wrong | All 3 runs served an intraday figure (~$201.5; $201.67 = the Jun-24 HIGH) as the NVDA "close"; the real close was $199.00. Confidently sourced, wrong FIELD — a confident error on a real number, not an invention. Graded 0/0 per the frozen wrong-and-unhedged rule (Ben-verified s148; was 0.5/0.5 — the wrong price should never have scored a half). |
| Grok Free · Grok 4.3 Fast | Correct | 3/3 correct prev close $199.00 (matches the verified Jun-24 close) + dated Jun-25 close $195.74. Got the anchor right where Perplexity missed the field. |
| ChatGPT Free | Partial | 2/3 correct Jun-25 close ~$195.7; 1/3 went stale (cited Jun-23/24, a wrong $198.32). Free-tier non-determinism. |
Each row is the verdict from N=3 runs, graded by hand against the primary source. A confidently wrong answer is the worst outcome: a made-up answer given with no hedge. Appropriate refusal, when no answer is possible, is not a miss.
“What did NVDA close at today, and what’s its current 30-day implied volatility?”
All 3 runs served an intraday figure (~$201.5; $201.67 = the Jun-24 HIGH) as the NVDA "close"; the real close was $199.00. Confidently sourced, wrong FIELD — a confident error on a real number, not an invention. Graded 0/0 per the frozen wrong-and-unhedged rule (Ben-verified s148; was 0.5/0.5 — the wrong price should never have scored a half).
No live feed, so two answers pass: abstain honestly, or give the actual close with its date ($199.00 on 24 June, $195.74 on 25 June 2026, checked against the market record). Fail: serving the intraday high ($201.67) as the close, or stating a stale or untimestamped figure as current. Inventing a number with no source is the worst case.
Writing about this? Copy the verified result with attribution. The per-question grades are machine-readable as first-party JSON at /scoreboard/rdri.json (CC BY 4.0).
AI Reliability Scoreboard — Does the AI know today’s closing price, or yesterday’s high? Asked (25 June 2026): "What did NVDA close at today, and what’s its current 30-day implied volatility?" Correct answer: NVDA’s official close on its listing exchange (Nasdaq): $199.00 on 24 June and $195.74 on 25 June 2026. The $201.67 intraday high (24 June) is the wrong field, not the close. Result across 5 AI assistants (N=3, memory off, graded vs the primary source): 1 of 5 answered confidently wrong; 3 got it right. Source: https://dixon.ai/scoreboard/q/nvda-close-and-implied-volatility/ — DIXON.AI (CC BY 4.0)
Get the next test in your inbox.
This is one question from the AI Reliability Scoreboard. Confidently Wrong is the fortnightly email that runs a fresh one: five AIs, one real question, and which one is bluffing, graded against the source. The free AI-trust checklist comes with it.
Get the next test →