Getting Real About AI Tools they’re Not Magic, They’re Just Pattern Machines

Everywhere you go, you’ll encounter content related to the fear of artificial intelligence (AI) taking over jobs. There are ads promising to create a complete blog post in five seconds or less. Then there’s your pal working in digital marketing posting creative graphics that appear as if they could have cost at least RM10,000 to design. People are buzzing about AI content generation tools and how they work. But let be honest often the way they are described is either too frightening or too good to be true.

There is a middle ground. AI is not SkyNet, but it’s also not just another toy or gimmick. If you are a small business, trying to generate copy for social media or attempting to learn how to use ChatGPT without sounding robotic yourself, here’s how all those applications really work, in plain English, based on my experiences working with these tools here in Malaysia.

So, what actually happens when you type a prompt?

Let’s look back at 5 years ago when you wanted a photo of a “nasi lemak burger.” You would have to hire a food photographer, purchase the props, prepare the food and take the photo yourself, or cook it on your own. When it was time to create a script for a video, you would just wait for inspiration to come at 2 a.m. until you got it. Today, you can simply enter “nasi lemak burger, neon lights, cinematic style” into AI image generation software, like Midjourney or DALL-E 3, and get four options in about two seconds. How do they know what nasi lemak looks like? They don’t know. That is the secret!

Generative AI models act like sponges that have absorbed trillions upon trillion of images from the internet. For example, when they see banana leaves, sambal or sliced cucumbers, they don’t know those are ingredients, but they do know when they see an image labelled “spicy sambal,” that image will typically contain dark red pixels. Therefore, when you request an image, there is no “imagination.” The AI uses numbers to statistically determine what pixel belongs next to which pixel. This holds true for text as well. An example of an AI text generation tool is ChatGPT. It does not generate “thoughts.” Instead it can predict what word goes next at a very high level. Therefore, you can say that it is like an autocomplete tool on several doses of steroids.

So, can these tools replace a real writer or designer?

Are they going to replace writers with artificial intelligence? Very simply: no. In a little more detail, they will replace the grunt work of writing. To illustrate, if somebody requests you to write a 1000-word essay about the history of Kuala Lumpur, the hardest part is the blank page, trying to find a starting point. Do you start in 1957? Do you start before that with Parameswara? The prompt-based AI content generation process removes the blank page problem, by giving the AI an instruction, i.e. “Write me an outline on the history of Kuala Lumpur.” In less than 10 seconds you have a basic outline: colonial, tin mining, independence. This is where the human comes back into play.

The AI’s finished product is likely to be something characterized as boring. The AI would describe KL as a busy, urbanized city. Not true, busy does not convey the level of chaos that occurs on Jalan Petaling during lunchtime! Only a human can provide the rasa (the feeling) of KL. Therefore, this is an example of collaboration between AI and human, allowing AI to do the structural, grammatical editing and draft product and allowing the human to do the finishing touches and seasoning it. Some software is also providing for automatic post content editing. For example, programmes like Grammarly or ProWritingAid scan your document and advise you of things such as excessive sentence length, or inappropriate comments. This helps you clean up your document so you can spend time being creative.

Moving pictures: The rise of video and voice

Words are easy to work with. Images are generally better, and videos? Videos were the most challenging type of media to produce. I can still remember producing a video for my friend’s store using Adobe Premiere—one hour of effort to produce a 30-second video. It was painful. Nowadays, AI content generation tools make it looks like child’s play: you write a blog post about “How To Change Your Oil” and throw that blog post’s URL into Pictory, it automatically identifies the best clips from the blog post, provides voice audio for each clip, and produces your TikTok video for you. Why is this important for the average person? Because people’s attention spans are decreasing.

No one even wants to read a 2,000-word article on Facebook anymore. People want to watch a video that lasts 60 seconds. To accomplish this, these AI video production tools check the text of your blog article for its key points using Natural Language Processing (NLP) software and then match those key points with appropriate stock footage. NLP is simply a way for computers to “learn” to read human language, like understanding that the word “sambal” means “hot” and the word “cendol” means “sweet.”

For Personalization & Optimization of Content application of this technology has taken on a whole new level. For example, if you have an apparel store, in the past you would run one advertisement for all of your customers. Now, by using AI, you can generate up to 100 versions of a single advertisement. The AI will change the shirt color in the ad based on what item your customer liked the most last week! This is an example of Multi-modal Content Generation—where we create content that combines text, images and data in order to create a message that speaks directly to one individual.

Is it expensive? Do I need a supercomputer?

TThis is fantastic news. No longer will you have to purchase an expensive PC to run powerful software. In the past, if you wished to operate powerful software, it was necessary to install a CD-ROM or download a large file from your hard drive before you could utilize the software. This is known as local deployment. Nowadays, the vast majority of applications are Cloud-based rather than local deployment based.

Cloud-based applications house their AI program(s) on the internet, and you can simply access the specific website that you want to use for your AI needs. For example: If you need a logo to create, use Canva’s AI capabilities, which are cloud-based. You need a script to write, use Claude or ChatGPT, which are also cloud-based. If you need to record a voice-over, use ElevenLabs, which is completely cloud-based. Most of the technological tools we use today will rely on API integration for AI tools. An API is merely a digital bridge that allows you to access Midjourney within your Discord application without having to own the actual Midjourney computer. Therefore, using the APIs allows you to use AI tools at a much lower cost or even for free.

But can I trust the numbers? How do I know if it works?

This is the part where people usually start selling you courses. They say, “Use AI, get 1 million views!” But how do you actually know if your AI content is good? You need to look at Content performance Analytics for Performance – fancy way of saying check your stats. When you write a post without AI you still check to see how many likes you got, right? That’s how you do it here too.

The best workflow is a loop…again. Thought of an idea thenused AI to create the first draft/image. After that, made it your own (added the Mamak shop language and the price in Ringgit). Used AI to create a prompt post for social media. Check your analytics again. If if you lose retention at 10 seconds, then you knew it was because your AI did not have a good hook (start back at Step 2).

IT’S CIRCULAR. It’s NOT “Set It and Forget It.” (By the way, there are also companies like BidaTech AI whose goal is to help manage this cycle so the company doesn’t go crazy from this cycle, but the intent is still the same; only the tools have improved). The tools are getting smarter, but they still lack common sense so You can play around with the tools and create kittens eating durian, shattered resignations in bed (Shakespeare style) and any other way that you want. Just remember that YOU are the one that has the authority; the AI is a tool.