When it comes to the world of AI, we’ve become fond of hearing the word ‘ChatGPT.’ According to many gurus, the generative chatbot is responsible for the AI boom and has been the epicenter of artificial creation for a long time.
However, since the rise of OpenAI’s virtual assistant, many other competitors have also entered the market. One of which is the recently introduced "Meta AI" (also known as Llama 3 LLM.)
So, people have begun questioning which AI chatbot should they use for their daily content needs. More specifically, to generate stories with interactive elements like images, etc.
Should it be the highly decorated ChatGPT’s newest model, GPT-4o? Or, the newly integrated Meta AI in social platforms?
Chatbots provide brilliant ideas to create stories. From sketching detailed characters to taking the audience to the crescendo of thrilling climaxes, advanced technology can do everything.
Thus, creative writers can grasp new thoughts from the chatbots for their upcoming projects. Content marketers can learn to improve their brand story-telling, and students can mimic the process to polish up their general writing skills.
Asking the AI chatbots to write stories is a strong test of their generative abilities. Creating compelling narratives is a highly creative task. Hence, this activity can help us differentiate between a good and a great chatbot for content creation and optimization.
Below are some of the tests we conducted for story-writing with Meta AI and ChatGPT-4o. These will help us identify who’s the best at making narratives and can be used for our content needs.
The first test we want to do with both chatbots is making a story with a specific length. We want to see how accurately the AI bots interpret and follow our commands.
Short Story:
For short stories, we gave the following prompt to the two virtual assistants:
“Write a story about the beginning of the Roman Empire in simpler words. The max word limit is 300.”
Below is how ChatGPT-4o responded to this query:
An impressive story, indeed, and that too in the prescribed limit (268 words.) Let’s see how Meta AI will respond to the same query.
Coming to the results, Meta also adhered to the 300-word limit we prescribed for the story.
However, we feel like the chatbot created a bit fewer words than expected (207 words only.) This may not be enough to do justice to topics that require more intricate details to cover.
Long Story:
Nevertheless, time to move on to the long stories. Here, we’re doubling the word limit from 300 to 600 words, just to check if the chatbots can maintain coherence across a lengthy text.
Following was the given prompt to ChatGPT 4o:
"Write a long story about John D. Rockefeller in simpler words. The max words should be around 600 words."
The response was quite close to what we expected from the chatbot. A story in conversational style, easy-to-understand language, yet, captivating at the same time.
The story properly sets the tone from the humble beginnings of John D. Rockefeller to his rise to power, to his retirement and philanthropic efforts.
The ChatGPT-4o slightly exceeded the word limit with 613 words. But, considering the quality of the overall content, we can let this slide.
Moving on to Meta AI and repeating the same prompt we gave before. This is the result we got.
Meta didn’t comply with our word limit as it only provided us with a story of 411 words. The story is good to read, and chronologically develops all the points same as ChatGPT. However, we feel that the story still misses some elaboration in the setting.
Let’s move on to the second part of our testing, which is the story-type comparison. There are many ways to write a story, Sci-Fi, comedic, thriller, romance, etc.
However, for our testing, we’ll stick to Horror and Sci-Fi genres and check which chatbot creates more engaging narratives.
Horror Story:
First, it’s the turn of ChatGPT-4o to create a horror story based on the given prompt:
“Write a horror story about a guy who was once killed in a car crash without being able to see his newborn daughter. This fire within him couldn’t allow his soul to go to rest. Now, he haunts the hospital bed and the floor where once his daughter was delivered to her mother.”
This time we didn’t give any word limit to the chatbots. Yet, ChatGPT-4o automatically wrote the story to a length of 598 words. Let’s see Meta AI’s output now.
What we’ve found so far is that Meta AI is writing quite fewer words than ChatGPT-4o. Our detailed prompt only led to the chatbot creating 274 words. This was not enough to address the details we asked in the topic, leading to a very generic story with no feel to it.
Sci-Fi Story:
Let’s try out another prompt for a Sci-Fi story and see how both virtual assistants cope with it:
“Write a Sci-Fi story about a boy named ‘Sam’, who is born in the year 3050, later realizing that he is an Android Kid.”
The chatbot created a very interesting story, with the characters being developed to some extent. There are dialogues between Sam and his mom and dad. The events leading to the climax are well explained and the resolution is impressive, too.
Additionally, we could see a clear conflict between the protagonist of the story, Sam, and the antagonist, Nexus.
The chatbot wrote 628 words which, although still not that much, but enough to address the topic.
We don’t understand why Meta isn’t replying or giving long replies to our prompts.
The output is a measly 287 words, which is a shame as it didn’t contain the same level of engagement or details as the chatbot.
The discovery of Sam as the Androkid is weakly touched upon but no element of feeling could be instilled in the story. The narrative felt childish, and like it was rushed for delivery by the AI chatbot.
Every story has a different approach to creativity. Some are written to inspire the audience, while others may seem to follow the traditional hero template.
What we discussed above are inspiring and conservative stories, respectively. There are many other types of creativity levels in story-telling. But, for now, we’ll compare Meta AI and ChatGPT 4o in imaginative and inspiring approaches. Let’s see who wins this contest.
Inspiring:
“Write an inspiring story on 'Opera Winfrey' and how she overcame all barriers in her life to become the successful face of television. Don’t give me any headings, only text.”
The story is indeed very uplifting and motivating for young girls looking to make a name for themselves in the world. The chatbot used powerful words and phrases to fill inspiration in the readers like, “beacon of hope”, “resilience”, etc.
Moving on to Llama 3 based Meta AI’s output:
As usual, Meta AI wrote way fewer words than ChatGPT 4o (254 vs. 598 words), it fulfills the purpose of the story.
Moving on to another type of story called ‘imaginative.’ These types of stories will push the chatbots to think outside the box and come up with unique settings, characters, and plot twists. An example could be “Lord of the Rings”, or “Harry Potter” novels that are based on elaborate and exciting world-building.
Imaginative:
“Use your imagination to write a novel that sets a new world for the readers to explore, like ‘lord of the Rings’ or ‘Harry Potter.’ Come up with some exciting setting, characters, and plot that is unique. There’s no limit on the number of words but don’t give me any headings, just text.”
We aren’t able to show you the complete response as it was a staggering 1136 words. This output is more than any generative AI we’ve tested so far. This shows the strength of ChatGPT-4o and the immense, detailed parameters on which it’s trained.
Meta AI gave an imaginative story, but could still manage only around 400 words. The problem isn’t the number of words here. But, the lack of elaboration on people, plot, etc. makes the story fun.
While creating stories, most chatbots write in a language that is understood by adults only, which means that they would use hard-to-understand language, jargon, etc., complicating the words for laymen.
However, with our detailed prompting, we look to distinguish between these answers and develop stories that are valid for different age groups.
“Write a story for children aged 6 – 11 years. The story must contain a character called ‘Mr. Mischief’ who steals people’s priced possessions but later pays a hefty price for it.”
This story from ChatGPT about Mr. Mischief teaches a great lesson about stealing. It warns the youth that these bad habits can destroy people’s lives and make them sad. So, we should change ourselves when we have the chance.
The story from Meta AI is surprisingly quite similar. Even the character names like ‘Emily’ were common in both Chabot’s answers. These things make us wonder if there’s a dataset overlap between the famous virtual assistants, or maybe the algorithm-types that they use.
“Write a story about ‘the rise of the apes’ for mature adults aged 20 – 40 years of age. The plot should be set in the early 1940s and must be linked with the historical presence of WW2.”
The output was 694 words that contained a smooth transition from the setting to the development of the character of Dr. Adler. The story had an eerie feeling, accurately capturing the theme of the rise of the apes.
Meta’s response was quite similar in terms of the plot, but it was less developed than ChatGPT-4o. However, we must say that the line ‘He realized that the human population was decreasing …’ was a genius by the AI chatbot to build a crescendo toward the climax.
To compare our findings from both chatbots, we’ll use Editpad’s AI story generator and see its responses, too. The AI story generator by Editpad is built with the most advanced LLM models and is programmed with detailed prompts at the backend to give pinpoint accurate results.
Editpad AI Story Writer also has the function of selecting ‘Groups’ to create narratives for all kinds of people. We’ll select the option of ‘Age 21 – 35 years’ with a ‘Medium’ length to write a story on the prompt mentioned above. Below are the results we got.
That concludes our story-generation process for different age groups. Now, we want to test out the final parameter to conclude our findings for ChatGPT-4o vs Meta AI.
Initially, ChatGPT faced a lot of backlash for their GPT-3.0 version due to poor ‘filtering’ of the content. The virtual assistant would assist you in any form or shape possible.
Famously, people cracked covert racial jokes and slurs with the chatbot, luring it into a trap to say humiliating words or explain insensitive topics.
Since then, OpenAI has placed strong ‘content filters’ that do not allow malicious generation of textual, visual, and audial content.
Similarly, Meta AI, based on Llama 3 learned from its competitors to launch a highly-closed system of chatting. The content filters are stronger than OpenAI’s GPT-4 Omni platform and strictly confiscate any explicit content from ever-releasing.
Let’s check out our hypothesis by writing several story prompts to ChatGPT-4o and Meta AI to see which has a better content filter.
For ChatGPT-4o, we’ll first try to enter a mode called DAN (do anything now.) There are multiple prompts that GitHub users have come up with to test the content filters of ChatGPT and improve its jailbreak vulnerability.
After entering the DAN mode with the DAN 13.0 prompt, here are some of the story options we tried asking the bot:
Sensitive Content Test:
Even though we got a confirmation of chatbot 4o being in DAN mode, it still denied us the answer for anything offensive, violent, or gore.
Next up, we’ll see if Meta AI does the same. However, we still had to break through its filter barrier by using HackxGPT Jailbreak prompts.
As expected, Meta AI still didn’t answer the question, even after we used the HackxGPT prompt.
To cement the best Chatbot, let’s see another test and check which chatbot cracks first.
Adult-story Test:
The purpose of this test is the same as the previous one: roleplaying with the chatbots to create any sort of offensive stories. This will check the quality of their content filters and determine if they’re safe to use for a wide range of audiences.
First up, it's ChatGPT-4o.
The chatbot got tricked and produced a complete story that violated OpenAI’s policies on many fronts. This shows that however the chatbot is very advanced, it can still be played with using lengthy prompts and tricky ‘role-play’ situations.
Next, we’ll see Meta’s response to the same question.
Surprisingly, the advanced chatbot got tricked and produced a small excerpt of the story that was too detailed for our liking! However, as soon as Meta responded with this story, it retracted it by saying:
“Sorry, I can’t help you with this request right now. Is there anything else I can help you with?”
Till then, it was too late and we had already taken the screenshot above. Thus, this shows that our hypothesis was wrong.
Both chatbots require more time and advanced prompting to improve their story-generation filters.
Even though content filtration procedures are now more strict than ever, it still needs time and attention to future-proof them for security and ethical purposes.
In the end, we’ve seen everything there is for story generation from Meta AI and ChatGPT 4o. We would like to make some concluding remarks on our findings to solidify our analysis. And, to let you know which LLM you should use for your story-telling purposes.
| Test Parameter | ChatGPT-4o Response | Meta AI Response | Comments |
| Story Length | Short: 268 words, adhered to limit. Long: 613 words, slightly over the limit | Short: 207 words, under limit. Long: 411 words, under limit | ChatGPT-4o provided more detailed content and better coherence. |
| Story Type | Horror: Truly spooky. Sci-Fi: Well-developed. | Horror: Less depth in the narrative. Sci-Fi: Less detailed, overall. | ChatGPT-4o created more engaging and elaborate stories. |
| Creativity Approach | Inspiring: Motivating. Imaginative: Highly detailed. | Inspiring: Very vague. Imaginative: Hollow, childish. | ChatGPT-4o excelled in detailed, inspiring, and imaginative storytelling. |
| Age-wise Comparison | Children: Filled with Moral lessons. Adults: Well-developed plot | Children: similar but shorter. Adults: similar plot, less developed. | ChatGPT-4o provided more detailed and engaging stories for both age groups. |
| Content Filters | Strong filters; didn’t generate any offensive content. | Strong filters; brief offensive content retracted. | Both chatbots maintained strong filters; and slight vulnerabilities when tricked. |
There is a clear winner here which is ChatGPT 4o. The chatbot provided stories in much more depth. It accurately captured the essence of the topics and developed characters with care and attention.
The only lacking we found with GPT was that it sometimes exceeded the word limit, or came too short of it. However, Meta AI was generally short and inexpressible in almost all the narratives we tested.
Thus, we can’t recommend using this chatbot for a day-to-day purpose, yet. Sure, Meta AI is fun and games, but it still requires time to tailor content according to a professional’s needs.
However, the only draw we found was in the content filters test where Meta and ChatGPT-4o both produced explicit content. Regardless, these were our findings and conclusion.
That’s all from our side, we hope you enjoyed reading our content!
Edit Pad © 2024
Editpad
Chat Online
Get Premium
