Featured | February 2, 2023
Could ChatGPT Replace Google?
Towards the end of 2022, OpenAI released its large language model, ChatGPT, to the public. It is built on top of the latest iteration of AI text-generation software GPT-3.5, with finer detail added thanks to human interaction.
With its chatbot-style interface, ChatGPT feels like a regular old IM conversation, just like what you’d have on Discord or Facebook Messenger. Except, you’re not talking to a human, you’re conversing with artificial intelligence.
ChatGPT can code, give article ideas, write poetry, and even emulate a Linux command line. Suffice to say, it’s pretty powerful. So much so that a portion of the internet has begun calling ChatGPT the “Google Killer”.
Now, Google is where us digital marketers earn a chunk of our livelihood, so this is a claim to be taken very seriously. So, in this piece, we’ll look at the timeline leading up to ChatGPTs release, its pros, cons, and use that to scrutinise the future of these information-hoarding platforms, and answer this important question.
What is ChatGPT?
For those that don’t know, ChatGPT is a large language model developed by OpenAI. OpenAI is a research and development company that deals exclusively with artificial intelligence and hopes to push the boundaries of what AI can achieve for mankind.
Using the platform is incredibly simple. Just visit the OpenAI website, create an account and in a few minutes, you’ll be able to start asking all sorts of questions to this cloud-based supercomputer.
What Makes ChatGPT So Unique?
If that explanation of ChatGPT didn’t get you excited, we don’t blame you. Chatbots like this have existed for many years now. Google Assistant and Amazon’s Alexa have been setting timers, giving us recipes and telling jokes for almost 10 years, so why is everyone suddenly so excited about (and scared of) ChatGPT? Here’s why:
ChatGPT Has Access To More Data Than Ever
There are currently 3 public iterations of GPT (Generative Pretrained Transformer). Each one has been trained on different sources (datasets) to improve its usability and knowledge. GPT-1 was trained on 7-10k unpublished books (exact figures are hard to determine); GPT-2 increased that to a whopping 40 GB of text data from roughly 8 million documents (including webpages); and GPT-3 uses a massive 175 billion parameters, totalling 800 GB of storage.
ChatGPT uses a brand new type of GPT that OpenAI refers to as “GPT-3.5”. We’re not yet sure how much information is contained within this new iteration, but it’d be safe to assume anywhere between 10-20x more data than that of GPT-3.
ChatGPT is Trained By Humans
On top of the enormous amount of data that ChatGPT can access, the underlying model has undergone a refinement process which OpenAI calls “Reinforcement Learning from Human Feedback”.
That process looks like this: a prompt is provided from a bunch of example questions, which is then answered by a human. That response is then fed to the AI as a possible outcome. Then, that same prompt is poised to the language model, which is given multiple opportunities to provide an answer. Technicians then rank the answers from best to worst, based on relevancy, information, harmfulness and other aspects.
This results in ChatGPT’s responses being much more accurate than ever before.
ChatGPT is User Friendly
Perhaps the most important factor that has caught the general public by storm is just how easy it is to use ChatGPT. There’s no need to clone a Github repository, self-host the AI dataset, or sign up to a third-party provider. Users simply create an OpenAI account, open the chat and away they go.
Positioning the tool as a chat also forces a more conversational tone to the prompts. Users can thank the AI, make jokes with it, and even refine their queries if the response they receive doesn’t quite match up with expectations.
Google and ChatGPT: Are They the Same?
Looking at the above, it’s clear that there are some stark similarities between Google and ChatGPT. Both platforms serve to show relevant information to users based on what they enter as a query, with that information being curated and ranked on multiple signals.
The main difference being that ChatGPT uses itself as its own resource, with that information barely being public knowledge. Google, however, uses a rich source of information to populate its search results: live websites. Let’s look at these differences in closer detail:
From our testing, ChatGPT is great at solving very specific issues and answering general knowledge questions that aren’t open to much interpretation.
Debugging code, explaining excel formulas, and breaking down proven concepts are all strong uses for ChatGPT when a simple answer is needed. It’s when complicated, delicate subjects are addressed where we start to see some of OpenAI’s filtering coming into action.
Ask ChatGPT an ethical question, and you can expect to see some form of follow-up disclaimer hereby revoking OpenAI of any responsibility for the information it gives back.
Since ChatGPT was released, many keen-eyed coders have been using it to write code unmonitored, giving answers to questions, most of which have arisen from Stack Overflow’s Q&A forum.
The problem is that ChatGPT’s responses can appear correct, while the reality is completely different. As a result, Stack Overflow has temporarily banned all ChatGPT-influenced answers to problems, stating that:
“The primary problem is that while the answers which ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to produce.”
While Google freely displays all the sources for its responses under featured snippets, OpenAI has held the dataset used to train ChatGPT under close grip. This is vastly different behaviour than what could be expected from a company claiming to be “Open”.
This means that the general public has no specific way of knowing what biases there are within datasets, or at which point the filtering and limitations take over. In its current form, these limitations are quite general. Try asking ChatGPT how to build a nuclear bomb or ask for its opinion on something, and you’ll see what we’re talking about.
The Importance of Options
Much like how an academic researcher needs to consider the opinions of others before making their own claims, sometimes, we want multiple answers to our questions to give us varying viewpoints on the subject at hand.
While users can refine their queries with ChatGPT by asking follow-up questions, there comes a point where the informed material can be manipulated by the user to read as however they’d like. Such clever wordplay has been enough to fool ChatGPT into thinking that its morality filter is switched off, allowing users to get answers to all sorts of problematic questions.
Google Isn’t Sleeping
Let’s not forget, Google has been heavily interested in the benefits of AI on its own platform for quite some time. Google’s own language model, BERT, was released back in 2018 as a method of training its search results to better understand how queries are written and understood.
There’s also project LaMDA (Language Model for Dialogue Applications), a seemingly direct competitor to ChatGPT that Google has been working on since 2017. While the project isn’t open to the public just yet, users can apply for testing via Google’s AI Test Kitchen.
LaMDA makes use of much the same principles that ChatGPT uses, except with the additional backing of Google’s immense catalogue of search data that it has gathered from years of existence.
Let’s not forget, Google has a reputation for starting and then promptly abandoning projects a few years after they’ve begun. Google’s cloud gaming platform, Stadia, was the latest victim to this, which was launched 3 years before work ceased, and had a rough user base of around 2 million users. If Google’s own chat bot fails to take off, it too could see an untimely demise.
Free is Free
Just recently, it was announced that ChatGPT had amassed 100 million monthly users just 2 months after its release. This makes it the fastest growing website of its kind; even faster than the most popular social media platforms.
More users equals more server load, which inevitably means sizable outages. This downtime has caused frustration, something that OpenAI is keen to rectify. To do that, OpenAI has announced a subscription model for the platform, dubbed “ChatGPT Plus”, which will be available at $20 a month (roughly £16). ChatGPT Plus promises access to the platform even during peak times, faster response generation, and priority access to new features as and when they come out.
Google on the other hand, is completely free to use, with the platform being funded via paid advertisements.
So, Is ChatGPT the Google Killer?
Much like any new technological release, there are still some kinks for ChatGPT to figure out while it finds its footing among the frenzied spotlight it has found itself in. Changes are happening thick and fast, with monetisation proving to be an aspect that OpenAI is keen to roll out in the coming months.
For now, it seems as though both tools could live harmoniously, with the ultimate choice being left up to users. Additionally, this competition has been brought directly to Google’s front door with the news that competing search engine, Bing, will be implementing an even faster version of ChatGPT into its search results pages. One thing’s for sure: AI is here to stay, and we’d better get used to it.
For more information on the latest technical marketing trends, have a read through our other blog posts. Or, if you want to speak directly with an expert, contact our team of talented digital marketers today.