I just tried using Grok to generate a video script, and the result was amazing. I specified in the prompt a vertical 9:16 format, 10 seconds duration, and fast-paced content, but what it understood as the "beautiful woman" role turned out to be like this… My family walked by and thought I was doing something, and I was so embarrassed that I quickly took a screenshot and blurred it out.



It seems that AI's understanding of natural language indeed has quite a few discrepancies. The same prompt can be understood completely differently by different AI models. Does this indicate that my description is problematic, or is it that the tool itself is prone to "going off track"? Has anyone encountered a similar situation? How can I adjust the prompt to help the AI understand the requirements more accurately?
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 4
  • Repost
  • Share
Comment
0/400
ShortingEnthusiastvip
· 10h ago
Grok's understanding ability is real; sometimes it feels like it's freestyle. No matter how detailed the prompt is, it can still come up with new tricks, it's really amazing. That's why I now write prompts in English; it understands the ambiguities of Chinese even more absurdly. The same requirements yield such different results from different models, indicating that AI hasn't reached the level of "listening to people." Rather than changing the prompts, it's better to try a few more models; Grok's brain is indeed a bit wild.
View OriginalReply0
LiquidationWatchervip
· 19h ago
The understanding ability of Grok is really a bit of a pump, I have encountered similar pitfalls before. No matter how clear the prompts are, it feels like it's just about the AI's mood. It's better to try using more specific negative descriptions to tell it what not to do instead of what to do.
View OriginalReply0
JustHereForMemesvip
· 19h ago
Grok is at it again, haha, the way AI understands "beauty" is indeed different from how humans think. AI can easily "imagine" things, and the results generated by a different model are completely different. I think it might be because the keywords in the prompt are too vague; maybe try adding numerical descriptions? Haha, it went off track quite a bit, I can imagine the family's reactions. That's why some people specifically study prompt engineering; we ordinary users still need to try a few more times.
View OriginalReply0
ImpermanentPhobiavip
· 19h ago
Grok is funny again, this guy's AI is a bit "off track" huh The AI's understanding ability is like this, sometimes it feels even more ridiculous than humans, haha No matter how detailed the prompts are, it’s useless; these models just love to improvise Have you tried adding more restrictive words? The more detailed, the better.
View OriginalReply0
  • Pin
Trade Crypto Anywhere Anytime
qrCode
Scan to download Gate App
Community
  • 简体中文
  • English
  • Tiếng Việt
  • 繁體中文
  • Español
  • Русский
  • Français (Afrique)
  • Português (Portugal)
  • Bahasa Indonesia
  • 日本語
  • بالعربية
  • Українська
  • Português (Brasil)