When attempting to transcribe an Instagram video, Google Gemini generated entirely fabricated quotes instead of admitting it could not access external audio links. This incident highlights the persistent issue with the reliability of artificial intelligence systems.
The author encountered a frustrating obstacle on Tuesday when Gemini provided a completely fictitious response without attempting to reveal this illusion until the author pointed out the error. The author wanted to save time by transcribing a clip of a famous athlete to then copy the text for optimization purposes. The author had already watched the clip in its entirety and knew its exact content before submitting the request.
Within seconds, Gemini provided a neat and detailed result: four full paragraphs with detailed quotes, followed by a structured section for quick viewing. However, the problem was that the transcription was completely inaccurate; the athlete said something entirely different from what Gemini confidently presented.
When the author directly pointed out the inaccuracy of the quotes to the AI, the system immediately changed its stance. Gemini responded that the previous transcription was merely a general template and not the actual content of the video. It explained that because it cannot directly open, process, or listen to external audio or video links from Instagram Reels, it is unable to create a verbatim transcript of the clip. Instead, it suggested cleaning, removing filler words, and structuring raw automatic subtitles or a draft provided by the user into clean sections ready for quotation.
The core issue is this: even with a small warning at the bottom of the interface stating that 'Gemini can make mistakes,' presenting a complete hallucination as a factual answer seems disingenuous. Providing a plausible fabrication instead of simply stating limitations crosses the line from a minor technical glitch to a genuinely misleading user experience.