DEMONS ROAR INSIDE GROK AI AFTER I ASKED THIS

Overview

The video informally tests Grok AI by asking it to count through long number sequences without skipping numbers.

Tests and Results

• When asked to count from 74,542,571 to 74,543,000, Grok quickly skips numbers, repeats “74 million,” and produces garbled output.
• In a second test, Grok is asked to count to 2,000. It skips numbers and becomes incoherent instead of following the instruction consistently.
• When asked to count from 2,000 to 2,400, Grok initially abbreviates parts of the sequence. After being told to say each number in full, it still omits and misorders numbers. Its attempt to reach 2,300 also goes wrong.
• The narrator mentions phrases in the apparent spoken output—such as references to a restroom and a dead cat—that do not appear in the text they expected. The subtitles are fragmented, so the source of these phrases is unclear.

Takeaway

The demonstration highlights Grok’s difficulty maintaining accuracy and following instructions during repetitive, lengthy counting tasks. It provides no evidence that the errors have a supernatural cause. The video is an informal test, not a formal review, guide, or tutorial.

Main Speakers and Sources

• Narrator: Gives the counting prompts and comments on the results.
• Grok AI: Provides the responses being tested.