Why is my GPT not following directions for word count?

ChatGPT
1

In the instruction window of my GPT, I have this written:

“Take any prompt and make it to be only 43 words long. Every output should be 43 words long, not 42 or 44 words long.”

And this is the chat transcript:

You: In a sprawling megacity, the neon-lit skyline blends with holographic advertisements. Distorted landscapes merge with cybernetic revelations. Towering megastructures loom over decrepit slums; augmented laborers tirelessly toil beneath polluted crimson skies. This dystopian tableau eerily captures the cyberpunk era’s harsh urban reality, starkly contrasting serene remnants of nature with the oppressive grip of technocratic dominance, encapsulating a ceaseless, metamorphic march of technological progress.

43 Words: In a neon-lit megacity, holographic ads blend into the skyline. Cybernetic elements distort landscapes. Megastructures overshadow slums. Augmented workers labor under red skies. This cyberpunk scene contrasts nature with technocratic control, showing relentless technological advancement.

You: That’s 35 words.

43 Words: In a neon-lit megacity, holographic ads and distorted landscapes merge. Towering structures loom over slums; augmented workers toil under polluted skies. This dystopian cyberpunk reality contrasts nature with technocratic dominance, symbolizing unending technological progression.

You: That’s 34 words.

43 Words: In a neon-lit megacity, holographic ads blend with the skyline. Cybernetic distortions merge with landscapes. Towering megastructures overshadow slums; augmented laborers toil under crimson skies. This dystopian cyberpunk era contrasts nature with technocratic control, encapsulating relentless technological progress.

You: That’s 38 words.

43 Words: In a sprawling megacity, neon lights and holograms blend with the skyline. Distorted landscapes and cybernetic revelations merge. Megastructures tower over slums; augmented laborers toil beneath polluted skies. This dystopian cyberpunk scene contrasts nature with technocratic dominance, symbolizing unceasing technological progress.

Surely, GPT can find one more word to make it a 43 word prompt? I’m using creator.nightcafe.studio to create art and it has a 43 word limit. Thanks!

2 Likes
2

While it’s a common question, it’s not a bug, so I’ll change the category.

I won’t attempt to explain this since I don’t know your background and level of understanding. It could lead to a lengthy discussion, which I’d rather avoid.

However, ChatGPT can provide valuable information and engage in discussions with you. :blush:

For example

You

Why are generative models not able to create completions with an exact word count? Help me to understand.

November 26, 2023
ChatGPT

Generative models like GPT-3 or GPT-4 are designed to produce natural language text that is coherent and contextually appropriate, rather than to adhere to strict constraints like an exact word count. Here are some reasons why it’s challenging for these models to create completions with an exact word count:

  1. Variable Length Responses: Natural language is inherently variable in length. The same idea can often be expressed in a few words or expanded into a detailed explanation. Models aim to generate text that is natural and fluent, which often means varying the length based on the content.

  2. Complexity of Language: Languages have complex structures and rules, and the length of a completion can depend on many factors, including grammar, style, and the specific information being conveyed. Generative models prioritize grammatical correctness and relevancy over word count.

  3. Predictive Nature: These models predict the next word in a sequence based on the previous text, aiming for the most likely or coherent continuation. They don’t have a mechanism to predict exactly how many words will be needed to complete a thought or sentence ahead of time.

  4. Lack of Explicit Counting Mechanism: Generative models typically don’t have a built-in way to count words and adjust the generation process accordingly. Implementing such a feature would require additional complexity and might compromise the quality or naturalness of the text.

  5. User Intent: Often, the exact word count isn’t the most crucial aspect of a user’s request. The model focuses on understanding and generating text based on intent and context rather than numerical constraints.

To produce text with an exact word count, additional post-processing steps or constraints would need to be applied, potentially sacrificing the natural flow and coherence of the text. For applications where word count is critical, custom solutions or manual editing might be necessary after the initial text generation.

1 Like