Share This Article
ChatGPT wrote the Bar Exam. ChatGPT 4 can replace some of the jobs. ChatGPT 4 outperforms its predecessor.
This is what has been making headlines in recent days.
It has been the talk of the town for the past two-three months. And I have not written anything about ChatGPT.
I think I’m getting FOMO?
So, here I am today, hoping for this trend and sharing my thoughts on ChatGPT and ChatGPT-4.
ChatGPT-4, the Open AI’s latest release. It has marked a significant milestone in the field of artificial intelligence, especially in natural language processing.
Impressive right? Let’s see if it’s impressive or still flawed.
GPT (Generative Pre-trained Transformers)
Yes, that’s the full form of GPT.
But, why are they called generative pre-trained transformers?
So, GPT, or the Generative Pre-trained Transformers, is a kind of deep learning model that generates human-like text.
Its common uses are:
- Answering questions
- Translating texts to other languages
- Generating code
- Summarizing text
- Generating code
- Generating stories, articles, and other content types.
AI existed before chat GPT. However, this AI revolution in natural language became possible with the introduction of these transformer models. This whole story started with the release of Google’s BERT in 2017.
But, how were things before GPT?
Let’s look back at the past.
Before GPT, text-generation-like tasks were performed with different deep learning models, such as recursive neural networks (RNNs) and long short-term memory neural networks (LSTMs).
They were good for short phrases and single-word suggestions, but they could not generate realistic longer content like articles.
Soon, OpenAI released a paper on “Improving language understanding by GPT.” This was GPT-1. In the same year, OpenAI published another paper titled “Language models are unsupervised multitask learners, for their latest model GPT-2.
The new GPT was obviously more advanced than GPT-1 and it often generated a couple of sentences. In 2019, it was considered state-of-the-art.
The next year, in 2020, OpenAI again published a paper titled “Language models are few-shot learners.” This was about their GPT-3 model.
It had 100 times more parameters than GPT-2 and it was trained on a larger text dataset for better model performance. This model continued to be improved with various variations, like GPT-3.5 and ChatGPT.
Soon, ChatGPT spread like wildfire, amusing the world with its power to generate human-like text. And in just two months, ChatGPT became the fastest-growing web application ever.
Now, I haven’t used GPT-1,2 or 3, but I have tried ChatGPT and ChatGPT-4. So here are my thoughts about them.
Coming from the content industry, and with all those human vs. AI comparisons going around, I couldn’t resist myself from trying out ChatGPT and understanding how it’s going to impact the content industry.
So, here is my opinion on ChatGPT
- I entered many queries and found *repetitions* in the ChatGPT answers.
- I looked for answers for similar keywords and phrases by rephrasing the questions. But all the output was pretty much similar. However, if you search for the same keywords on Google, every output will be different. I mean the intent of the output will differ like marketing intent, informational intent, and so on.
- I asked ChatGPT to write a 500 words blog on a certain topic and the blog ended up with 56% plag. I mean some of the paragraphs were the exact same word-by-word from the source.
- Many of the links provided by ChatGPT were dead.
All of these points make it pretty clear that no matter how you train the AI, it won’t be able to add context like we humans do.
But here I also feel that ChatGPT can be helpful in various other segments like video scripting, writing code, getting content ideas, or finding errors in the code. But there will always be a human beating it.
I also tried asking more questions to ChatGPT. Just for fun, ? and here are the results.
I asked ChatGPT the same question- Human vs. AI content, which one is better?
I was fascinated with the result.
Here are some of the funny snippets from my other search queries.
Let’s talk about ChatGPT-4.
What’s New in ChatGPT-4?
ChatGPT-4 is the new and more advanced version of its predecessor. It is an expert on various subjects. Now, the ChatGPT can describe images, making it impressive. But is it flawless?
Let’s find out.
- The ChatGPT-4 has learned to be more precise.
- It’s accuracy has increased (but not 100%)
- One good thing is that it can describe images with striking details (i saw that someone clicked his open refrigerator pic and asked ChatGPT-4 to suggest a recipe based on the available ingredients, and it gave a detailed recipe to it.)
- It can ace standardize tests. (you all must have heard ChatGPT qualifying the Bar exam)
But, this new bot is still making stuff up.
This is one issue that all chatbots have. They don’t have an understanding of what’s true and what’s not, sometimes they generate completely false text. It is called “hallucination.” Giving dead links is again can be seen here.
Moreover, if we talk about technical challenges, undoubtedly, GPT-4 performs better than its older version.
In the end, I’ll again say the same thing, AI cannot beat humans, but, humans can leverage these AIs to make their tasks easier and more achievable for themselves.
I would recommend watching this 1 min video from NasDaily. It has explained the role of AI in the best way possible
https://www.instagram.com/reel/CqIiwqvAQAN
For more such interesting articles, subscribe to our newsletter.
