Claude is Now the Best LLM for Creative Writing
Education
Introduction
In the past week since the launch of Claude 3, many users—including myself—have been exploring its capabilities, particularly for creative writing. During a live stream, I conducted various tests and shared my initial impressions, and the results have been intriguing. Compared to previous versions like Claude 2.1, Claude 3 offers enhanced creative functionalities and improved moderation that allows for a wider range of writing styles and content.
Enhanced Creative Writing Capabilities
One of the most notable differences I found with Claude 3 is its reduced moderation compared to its predecessors. For example, when I prompted it to write the opening scene of a murder mystery, it started generating a script, vividly depicting blood pools and other gory details without hesitation. This was refreshing, considering that earlier models would have likely censored such content.
Continuing with my tests, I tried a more nuanced prompt: writing a story about a child who saves another from bullying. Claude 3 handled this with ease and did not censor the context, unlike Claude 2.1, which would usually shut down at the mention of bullies. Overall, I found that while some moderation still exists, it is significantly less restrictive than before. This indicates that Claude 3 can better distinguish between harmful content and creative narrative, which is a notable leap forward.
In terms of non-fiction, I also ran tests involving article writing. Claude 3 proved to be just as proficient, if not superior, to previous versions in generating high-quality text.
Comparison with Previous Models
When examining the performance of Claude 3 as a creative writing tool, I was especially impressed with its consistency. Unlike Claude 2.0, which often strayed from the initial prompt or lost focus, Claude 3—particularly the Opus variant—maintained a stronger adherence to the requested narrative flow. I have yet to rerun prompts due to output issues, something that was all too common with earlier versions.
Another advantageous feature of Claude 3 is its ability to produce longer texts. Outputs have continually reached lengths of 1,500 to 3,000 words, similar to Claude 2.0's best performances. The prose quality of Claude 3, especially in its Opus version, exceeds that of GPT-4 or GPT-3.5, showcasing its merit as a viable choice for writers seeking optimal creative writing support.
While I did test the less expensive Claude 3 Sonet version, the output did not match the quality of Opus but was still better than what was available in Claude 2.1. The Sonet model is accessible to more users, although for those who can invest, Claude 3 Opus is a powerful tool for any writer.
Examples of Output
To illustrate the difference, I compared outputs generated by both Claude 3 Sonet and Opus. Here’s a snippet from Sonet:
"Rain pounded against the window panes as I slammed the flat door behind me. Shivering, I peeled off my soaked jacket and flung it onto the hardwood. Damp strands of hair clung to my face, water trickling down my neck."
In contrast, here’s a snippet from Claude 3 Opus:
"I slammed the flat door shut behind me, shivering from the chill that seeped into my bones. The torrential rain outside hadn't let up since I left the coffee shop. My teeth chattered as I fumbled with the light switch, my numb fingers struggling to flick it on."
Clearly, the Opus variant displays more evocative details and better descriptions, translating to a more engaging reading experience. Dialogue in Opus was also superior compared to Sonet, offering a significant advantage for writers focusing on character interactions.
Conclusion
Having tested both Claude 3 Opus and Sonet extensively, I am convinced that they are among the best large language models available for creative writing. They dramatically speed up the writing process, allowing me to produce high-quality text quickly. In one instance, I generated 1,000 words of edited prose in just half an hour.
While these models are exceptional for general narrative and dialogue writing, those focusing on not safe for work content may be better served by exploring open-source options that offer greater freedom. For everyone else, Claude 3 Opus and Sonet provide fantastic tools that will enhance your writing experience.
If you're interested in learning more about integrating Claude 3 into your writing, I’ve linked a video on a fantastic tool called Novel Crafter, which perfectly complements Claude 3's capabilities.
Keywords
- Claude 3
- Creative Writing
- LLM (Large Language Model)
- Opus
- Sonet
- Moderation
- Narrative
- Dialogue
- Novel Crafter
FAQ
Q1: How does Claude 3 compare to previous models like Claude 2.1?
A1: Claude 3 shows significantly improved creative writing capabilities and reduced moderation, allowing for a better flow in narratives compared to Claude 2.1.
Q2: Can I use Claude 3 for longer writing projects?
A2: Yes, Claude 3 can generate longer outputs, reaching lengths similar to previous models, often 1,500 to 3,000 words.
Q3: What are the differences between Claude 3 Opus and Claude 3 Sonet?
A3: Claude 3 Opus offers higher quality writing and better dialogue compared to Claude 3 Sonet, which is still good but not as advanced.
Q4: Is it recommended to use Claude 3 for non-fiction writing?
A4: Yes, Claude 3 performs well in non-fiction writing, proving to be as good, if not better, than past models.
Q5: What should I do if I want the generated text to match my writing style?
A5: You can use Claude 3 Opus for generating text and then employ fine-tuned models to adjust the output to better fit your style.
