Forum Discussion
Microsoft Copilot cant spell
Hi All ... Can I get some help here, perhaps I am prompting incorrectly. Very frequently Copilot will misspell words even if its provided in the prompt. I may request a new image using the word OPERATE and it will misspell it as OPERRAT, or using the image below a Lenovo laptop misspelled as LEYOZVO. Why is this occurring and how do I get Copilot to fix it?
6 Replies
I am having a problem with this too. I even give it explicit instructions, spelling words exactly how they should appear (correctly) and it still screws it up.
These responses (or perhaps better said "excuses") are ridiculous.
AI can generate text to display to the screen. Yes.
AI knows the sequence of the characters when displaying text to the screen. Yes.
So how is it that an AI graphics generator cannot be programmed to generate the sequence of characters as blocks of pixels side by side in the proper order?
Still baffles me.
Karim111 Hey, unfortunately the way the generative AI works is to look for patterns in images - it doesn't actually 'know' the text it's creating. It just sees the letters and words as patterns of pixels. That's why it struggles with text (and the number of fingers on a person's hand 😆).
Things are improving and different models do it better, but it's still a basic limitation. I tend to avoid trying to get it to do any sort of accurate text creation.
Interesting that hundreds of millions of dollars are invested in AI data centers across the world and the systems are straining our entire utility grid but it cannot provide simple accurate spelling.... asinine!
Text in AI-generated images can be misspelled for several reasons:
1. Training Data Limitations: The datasets used to train AI models often lack a strong emphasis on correct spelling, leading to errors.
2. Tokenization Methods: AI models may tokenize text based on whole words or phrases, which complicates the task of spelling individual letters accurately.
3. Visual Prioritization: AI models like DALL-E (used by Copilot) prioritize visual elements over textual accuracy, making text generation more challenging.
4. Complexity of Text: Text symbols are seen as combinations of lines and shapes, and the variety of fonts and styles makes it difficult for the model to reproduce text accurately.
These factors contribute to the occasional misspellings you see in AI-generated images. If you need precise text, it's often best to add or correct text using graphic design tools after the image is generated.
Hope this helps.
