The Ideal LLM for Content Writing
- Tarasekhar Padhy
- Oct 16, 2024
- 3 min read
(Disclaimer: Profanity.)
Over the past couple of years, I’ve had the privilege (and pain) of using many LLMs for content creation. While each of them had distinct advantages, none of them were perfect.
When I use ChatGPT, for instance, I can’t help but wish that it had a larger context window and better reasoning capabilities. With Claude 3.5 Sonnet, I want internet browsing capabilities.
The problem was each of the AI-powered chatbots I used lacked something that prevented it from being the LLM for content creation. After hours of overthinking, I couldn’t help but wonder if there ever can be a perfect LLM for content writers.
So, let’s begin by looking at the features of an imaginary LLM’s features that will make it the all-in-one content writing machine.
Features of an ideal content writing LLM
Large context window: It has to be around 512K. I am aware that it’s a bit too much and I am being quite optimistic here, but it is important for going through long reports and even eBooks with hundreds of pages. Moreover, lengthy conversations won’t veer off into an unproductive tangent and the likelihood of hallucinations will decrease.
Customizability for non-coders: Right now it is possible to connect AI tools with other software via APIs. But this limits its usage by content writers. There are many workflow upgrade ideas such as connecting Notion with GPT-4o which I can’t do. I should have opted for computer science engineering instead of outdated mechanical.
Better information retrieval: Claude 3.5 Sonnet and Gemini 1.5 are the best at this in the market. ChatGPT is like an inbred with Parkinson’s when it comes to analyzing documents, files, etc., and extracting context specific information for you. Gemini 1.5 can even analyze hour-long podcast videos, directly from YouTube (via Notebook LM)!
Updated knowledge base: The only LLM with a respectable enough global knowledge base is GPT-4o. Everyone else is just pushing out barebone chatbots which do nothing for most regular content writers. It sucks donkey balls because the ones with great information retrieval capabilities have a limited database.
The ideal LLM doesn’t exist
I am completely aware that packaging all the aforementioned features into one tool is easier said than done. I also know that all of the nerds in San Francisco and other tech capitals of the world are smarter than me, and that’s putting it lightly.
Furthermore, there are also certain properties that work against each other. For instance, a model with good information retrieval capabilities probably fares poorly in creative tasks. And chatbots with large knowledge bases will usually give longer replies, even to basic questions.
However, there are definitely some low-hanging fruits for various proprietary LLM providers across the AI market.
I think the nerds at OpenAI should at least double the context window of their flagship model to 256K tokens. The incels at Anthropic should give their model internet browsing capabilities to take the fight to Microsoft and so on.
Whether these upgrades come in the future or not, one thing is definitely certain, they will surely get more affordable. This will level the playing field where the only differential will be your prompting skills and understanding of the action item for which you are using AI.
Conclusion: Dress for the occasion
For short-form thought-leadership content, Claude 3.5 Sonnet is best. When it comes to conducting in-depth research from different academic, literary, and video sources, nothing beats Gemini 1.5.
And all of the remaining tasks, ChatGPT still reigns supreme.
Keep in mind that most of the general tasks, like generating title suggestions on common topics or proofreading text, can be done with all the LLMs. It is crucial to learn the specific strengths and weaknesses of each proprietary model available to make the most out of them.
—
Next Chapter: How to Build a Custom Content Writing GPT
Previous Chapter: ChatGPT Canvas and Claude Artifact: Good Assistants
Index (with Prologue): Content Writing With AI