The 2 basic workflows for ai-assisted content generation

Table of Contents

The two basic kinds of content creation workflows when working with LLMs are these:

Semi-automatic

LLM -> editor -> publication

Fully automatic

LLM -> publication

Of course there can be various -pre, -in between and -post processing steps by code in either.

But here I focus on whether a human is involved in the content creation workflow or not.

First, let’s have a look at the advantages of either approach:

Advantages #

Fully Automatic Workflow	Semi-Automatic Workflow
Speed - Processes can be completed rapidly without human intervention.	Quality Control - Human editors can ensure higher accuracy and quality by intervening when necessary.
Consistency - Outputs are uniform without any subjective variation.	Flexibility - Adaptations can be made on-the-fly based on editor insights or external feedback.
Scalability - Easy to scale operations to handle large volumes of tasks without additional human resource costs.	Error Correction - Human insight allows for immediate error identification and correction, which might be overlooked by automated systems.
Cost Efficiency - Reduces labor costs associated with manual tasks over long periods.	Critical Thinking - Humans can apply nuanced judgment that is difficult for automated systems to replicate, especially in complex scenarios.
24/7 Availability - Can operate continuously without breaks or downtime needed by humans.	Process Improvement- Human intervention secures immediate feedback on workflow results, improving processes over time.

And then, what are the disadvantages?

Disadvantages #

Fully Automatic Workflow	Semi-Automatic Workflow
Lack of Flexibility - Struggles with handling exceptions or unexpected situations that are not pre-programmed.	Slower Processes - Involvement of human editors can slow down the workflow compared to fully automated systems.
Limited Understanding - May not interpret complex or nuanced information as effectively as a human.	Higher Costs - Requires ongoing investment in human resources, which can be more expensive than fully automated solutions.
Dependency on Data Quality - Performance heavily reliant on the quality and breadth of the input data.	Inconsistency - Human intervention might introduce variability in the output quality depending on the individual editor’s skills and judgments.
No Personal Touch - Unable to provide the human-like interactions or decisions that might be necessary in some fields like customer service.	Scalability Issues - Scaling up involves additional training and hiring, which can be costly and time-consuming.
Over-reliance on Technology - Can lead to reduced skills among staff and an over-dependence on automated systems.	Fatigue and Errors - Human editors can make mistakes, especially when handling repetitive tasks or working under pressure.

As always you cannot have it all. There are trade offs to both techniques.

In reality it is good to switch back and forth between to the two workflow types especially during the development phase to reap the awards and insights of both.

It goes without saying that some content creation tasks are more suited for full automatization while others require human oversight.