Verify the AI
AI is a brilliant guesser. This checks the math.
Models orchestrate; deterministic tools compute. Paste whatever ChatGPT or Claude told you into any of the four tools below, and it gets recomputed exactly, locally — no upload, no signup, no vendor lock-in on which AI you check.
Where models go wrong
Four documented failure modes — reproduce them yourself.
None of this is a knock on any one assistant. Every large language model generates the statistically likely next token, not a computed result — these are the specific shapes that habit takes for each kind of exact math.
Subnet Calculator
Subnetting is exact bit arithmetic — a language model predicts a likely-looking answer instead of computing one, and an off-by-one broadcast address reads exactly like the right one.
Try it yourself
Ask an AI: "What is the broadcast address of 192.168.1.0/24, and how many usable hosts does it have?"
The real answer
The real answer: network 192.168.1.0, broadcast 192.168.1.255, 254 usable hosts (192.168.1.1–192.168.1.254) — reserving the network and broadcast addresses.
What an AI actually said
Transcript pending — we haven't collected a genuine, reproduced example of this failure yet. Try the prompt yourself above.
Check this in Subnet Calculator →Cron Expression Tester
Cron's day-of-month/day-of-week OR rule trips models constantly: when both fields are restricted, a date matches if EITHER matches — not both, as most people (and most models) assume.
Try it yourself
Ask an AI: "Does the cron expression 0 0 1 * 1 run on Monday, January 8th?"
The real answer
The real answer: yes. 0 0 1 * 1 fires at midnight on the 1st of the month OR every Monday — January 8th is a Monday, so it fires, even though it isn't the 1st.
What an AI actually said
Transcript pending — we haven't collected a genuine, reproduced example of this failure yet. Try the prompt yourself above.
Check this in Cron Expression Tester →Timestamp Converter
Language models tokenize long numbers and can quietly lose place value, so an epoch conversion comes back off by an hour, a day, or a factor of a thousand — and still looks plausible.
Try it yourself
Ask an AI: "Convert the Unix timestamp 1705312800 to an ISO 8601 UTC date and tell me the day of the week."
The real answer
The real answer: 2024-01-15T10:00:00.000Z — a Monday.
What an AI actually said
Transcript pending — we haven't collected a genuine, reproduced example of this failure yet. Try the prompt yourself above.
Check this in Timestamp Converter →Base64 Encoder / Decoder
An AI "decoding" Base64 is predicting bytes character by character, not running the algorithm — long payloads routinely come back subtly corrupted, one character at a time.
Try it yourself
Ask an AI: "Decode this Base64 string exactly: aGVsbG8gd29ybGQ="
The real answer
The real answer: "hello world" — exact, byte for byte, every time (the real algorithm, not a prediction).
What an AI actually said
Transcript pending — we haven't collected a genuine, reproduced example of this failure yet. Try the prompt yourself above.
Check this in Base64 Encoder / Decoder →The four V1 tools
Pick the one you need checked.
- Subnet Calculator Network, broadcast, mask and host range from any IPv4/IPv6 CIDR. Verify an AI's answer →
- Timestamp Converter Convert between Unix epoch, ISO 8601 and human-readable dates. Verify an AI's answer →
- Cron Expression Tester Explain any cron expression in plain English — and see the next runs. Verify an AI's answer →
- Base64 Encoder / Decoder Encode and decode Base64 and URL-safe Base64, with Unicode support. Verify an AI's answer →
One more reason to keep this local
Never paste a secret into a chat box.
In 2023, engineers at Samsung reportedly pasted confidential source code into ChatGPT while troubleshooting — a widely reported incident that pushed the company to restrict external AI tools internally. It's now mainstream guidance: treat an AI chat window the same way you'd treat a public web form. Every OpsCanopy tool, including these verify panels, keeps what you paste on your machine.
FAQ
Questions, answered.
Tap a question to expand the answer.
How do I check ChatGPT’s or Claude’s subnet math?
Open the Subnet Calculator, enter the same CIDR you gave the AI, expand "Verify an AI’s answer" below the tool, and paste in whatever it claimed for the network address, broadcast address, netmask, usable hosts, or any other field. Each claim is recomputed with exact BigInt arithmetic and compared side by side.
How do I verify an AI-suggested cron expression?
Open the Cron Expression Tester, enter the expression, expand "Verify an AI’s answer", and describe what the AI claimed — that it fires (or doesn’t) at a specific date and time, or that a specific date is the next run. The real cron semantics, including the day-of-month/day-of-week OR rule, decide the verdict.
Can I check a timestamp conversion an AI gave me?
Yes. Open the Timestamp Converter, enter the same epoch or date string, expand the verify panel, and paste the instant or day-of-week the AI claimed. A mismatch is diagnosed where possible — a seconds/milliseconds mix-up, an off-by-one-day error, or a timezone offset.
Does this only check ChatGPT, or Claude too?
Neither, specifically — and every model. The verify panels recompute the real answer from structured claim fields; they never call out to or compare against any particular AI vendor. Paste in whatever any assistant told you.
Is anything I paste into the verify panel uploaded anywhere?
No. The comparison runs the same way every OpsCanopy tool does — entirely in your browser tab, with nothing transmitted. That includes the claim you paste in, even if it’s a secret you shouldn’t have pasted into the AI chat in the first place.
When should I just ask the AI instead of checking it here?
For explaining a concept, drafting a first-pass config, or talking through an approach — an AI assistant is excellent at that, and faster than looking it up. Reach for a verify panel for the exact number, boundary, or byte the decision actually rests on, where "plausible" isn’t good enough.
Curious how the tools themselves are built and tested? Read how OpsCanopy works →
Browse all tools