Ever since GPT-3 was released, there has been an influx of tools with AI in their name that can help you write the next blog post, advert, product description, email, or anything else you need content-wise. The AI revolution is truly underway but should you use an AI writing assistant tool?
In this blog post, we'll cover the pros and cons of AI writing assistant tools and why we think that there is no better step you can take in your AI journey than learning to utilize the raw technologies for your needs, and how AI writing assistant tools are unnecessary middleman dampening the quality of output for your business and brand.
As we try to showcase and make it as easy as possible within Riku, building out a prompt for an AI content generator is a relatively simple task. In short, all that is needed is an instruction, a few examples, and ensuring the toggles of the settings are set correctly to give you a quality output. Of course, one of the main downsides of prompt building is that you are limited by the token limitations of that AI engine. For most, this is 2048 tokens (approx 8,200 characters) so you can only provide examples to cover a few different business use cases.
If I am building out a prompt to use in an AI writing assistant tool, I want the prompt to be able to provide a reasonable output for the end-user. The end-user might have a plumbing business, be a dentist, or run an ecommerce store and my goal as an AI writing assistant tool creator is to ensure an output relevant to that business so the prompt I will build for my tool will only focus on the surface level of a multitude of different businesses to do this. Ever wondered why when using an AI writing assistant tool you hit generate and get the output and are left with a 'meh' feeling. The content generated is technically correct but it is pretty average and has no depth to it. This is the design of AI writing assistant tools. It is working as intended by the creator.
Of course, not all AI writing assistant tools are using just prompts in their products. Some will use fine-tuning which eliminates some of the token issues mentioned previously but it still does not cut out the problem of providing only surface-level outputs. A larger dataset of generic content not specific to an individual business will still only output generic content not specific to an individual business.
One of the ways to get deeper, higher-quality outputs from AI is to go to the source. Building out your own prompts is incredibly easy and you can tailor them specifically to your brand and situation. If I am running a dental practice in Oklahoma focusing on dentures and whitening then I want the AI to know this in the prompt and by knowing this, I'm going to get a more relevant output.
When using an AI writing assistant tool, you may have the opportunity to enter a few inputs with small details about your brands. These are often limited by a character count and just tack on to the end of a pre-designed prompt so whilst you do provide a little relevancy, in the grand picture of the prompt or fine-tune that has been designed, it is only going to be to a surface level. Nothing compares to building out your own prompts or fine-tunes and by doing that you can craft it exactly to your individual needs.
Think of it another way, if I am using a prompt made up of 10 examples and the 10 examples are all random businesses not in my sector when I come to use that prompt, I'm going to get a surface-level output but nothing specific. If the prompt has been created specifically for my brand with 10 ultra-relevant pieces of content then the chances are that the AI will have a deeper understanding of my business and be able to build on that knowledge to go deeper and provide an output that is going to be super specific for my needs.
AI writing assistant tools are well aware of this problem. Some have tried to solve it by offering the ability of users to create custom frameworks. Custom frameworks are effectively giving you half the tools you need to build out a prompt. You are given a canvas to add some examples and display a pattern but you aren't shown what technology is being used or able to really go deep in tweaking the finer settings like temperature, frequency penalty, etc.
One of the ways I like to look at this is thinking of it like you are on a diet. If you are on a diet, you want to know every single little thing that goes into your body. Eating out at a restaurant saves you time and effort in actually preparing your meals but you don't know how much salt they are using, you don't know the quality of the ingredients, and are ultimately kept in the dark, custom frameworks are a similar process.
AI writing assistant tools have a vested interest in not giving you the full feature set to build and edit your own prompts because if you did, you'd just go to the raw AI large language model provider and do it yourself, it makes them redundant. If you are using such tools and find yourself using frameworks most of the time, then chances are you are ready to take the red pill and jump headfirst into the world of AI for yourself and have outgrown these tools.
You will save time by using an AI writing assistant tool as the UX is often awesome and it is easy to quickly generate multiple things for your needs. The quality might be a bit hit and miss but in terms of writing manually or using such a tool for the first time, you can be quite happy with what you see and the time that is saved.
The question is also then, are you willing to invest a little bit of time in setting up your own prompts for an AI output that is a lot deeper and higher quality for your needs or are you content with the surface-level generations provided by an AI writing assistant tool? People value different things and some users will not want to invest their time into learning the raw AI, that's fine and in such cases AI writing assistant tools can be awesome for your needs.
Here at Riku, we're obviously big advocators for those who do want to take that plunge into learning a completely new and totally valuable skill for the next 20 years and think that such an investment will give you a new superpower and open up many doors for you in the future. The choice is ultimately yours but if you made it this far, you must be pretty curious? Right :-).
Riku.AI is a platform that brings all of the best large language models for both text and image AI under one roof for you to experiment, learn and build with. Our community and environment make exploring and diving deep into AI a real pleasure for both technical and non-technical users.
