What is ChatGPT?
It’s essentially a conversational chatbot that is publicly available, which anyone can access by creating a free account. It’s powered by an AI technology called GPT-3, a type of large language model, which is a machine learning tool that has been trained on a huge dataset of digital copy with human supervision that grants it reward points for certain results. In other words, its upbringing has been similar to that of a human toddler. It was released to the public on 30 November 2022.
The project was created and developed by a research company called OpenAI and drew investment from Microsoft and others. The code is not open-source. There is an open-source alternative called PaLM + RLHF created by Philip Wang, but it hasn’t been trained yet.
Why is it growing so fast?
Before ChatGPT arrived, there were some preliminary steps into the field by others. Microsoft’s Tay was released in 2016 as a Twitter user. Within 1 day, it began tweeting deeply offensive replies and was shut down. Meta released Blenderbot 3, which was inaccurate and dismissed. Google’s LaMDA was never released to the public due to reputational risk, however, an engineer working on the project was suspended because he published a claim that it had become self-aware.
OpenAI’s company valuation has shot up from $260m to $20b in less than a year.
The technology has shocked the world, with schools warning kids not to use it to automatically generate their assignments, because it’s cheating. Some students are even using additional online tools to disguise the work and prevent their teachers from discovering ChatGPT’s involvement.
The meteoric rise has been so apparent, that Google’s top management has called a “code red”, signalling that this could be an existential threat to the leading search company.
Perhaps the human supervision ChatGPT received in training can account for its polite and inoffensive nature, allowing it to enter the public space and thrive where predecessors had failed. It’s been raised well.
Elon Musk, who needs little introduction and happens to be one of the founders of OpenAI, has said “ChatGPT is scary good. We are not far from dangerously strong AI”.
Is it a threat to Google?
At the moment, Google commands an 83% share of search traffic on the web. Microsoft Bing, the next largest search engine, only 9%.
The internet has sprung many soothsayers claiming with certainty that ChatGPT is a better question answering service than Google.
Why would it be a threat?
- Microsoft will be integrating ChatGPT’s technology into their search engine in March of this year. They’ve been working on doing this since 2019. Unlike Google’s BERT AI, which is only used to help rank the relevance of search results, Microsoft are reportedly going to use this tech to make their results more conversational. OpenAI is also currently working on GPT-4, an improved version of the AI system that powers ChatGPT. They’re also integrating it into other products in markets they share with Google, such as email and documents.
- There are many types of search queries that might be more aptly answered in a conversational style, than by a list of links. Examples might include “How do I cook a perfect steak?”, “What are the side effects of taking medication XYZ?”
- A cleaner, less ad-ridden experience for the end-user. Google’s & Bing’s main revenue streams are ads from the search results though. But Microsoft does have other revenue streams, which could allow them to compete harder in this space.
- It’s new and exciting! People like to feel like they’re stepping into the future.
Why wouldn’t it be a threat?
- At the moment, ChatGPT is not updated in real time. The information it provides is out of date (dated up to 2021 at the moment).
- It’s not accurate. It can quite impressively offer solutions to some even quite advanced maths or programming problems correctly, however it is due to luck and close adherence to the material on which it was trained. It is not able to actually perform calculations, or think critically.
- Certain queries are just better answered by a search engine. Examples include: “Weather forecast for tomorrow”, “What’s the population of Sydney”, “Directions to XYZ”, “list of trustworthy attorneys specialising in XYZ”.
- It’s computationally and financially very expensive. ChatGPT now costs about $100k per day to run using Microsoft’s Azure server infrastructure, for the million or so users currently signed on. This kind of AI works by scanning through large amounts of text and making exponential numbers of comparisons between words or phrases. It takes huge thinking power to do these tasks at scale, for example, for a search engine. Google’s BERT is used on only around 10% of queries.
- Google has been working on AI language models and chatbots for years. A big move from Microsoft will likely prompt Google to quickly release their own versions.
What CAN, and what CAN’T it do?
ChatGPT was not trained for specific tasks. It was trained with broad and general knowledge, and therefore is capable of performing tasks it hasn’t specifically learned how to do. A Stanford student published that Google’s AI called BERT, which is a similar natural language learning model, can teach itself universal grammatical principles that apply to different languages, and as a result learn to translate.
Built on top of InstructGPT, ChatGPT can take specific instructions and generate unique content based on the provided guidelines.
It can remember what you said earlier in a conversation, making it quite the charming conversationalist.
Can/should I use it to generate and publish unique content?
The short answer is no.
For the following reasons, it’s not practical to directly use ChatGPT to write your content for you:
- It’s not factually accurate. ChatGPT does not even aim for accuracy of information. It’s just predicting what words it should write next, based on its training material and what you asked it.
- It’s been deliberately designed to be biassed and limited. Its training makes it avoid hate speech, teaching you how to kill people, or generally being negative or dark-spirited.
- It’s completely unaware of recent, or real-time events. It’s incapable of looking anything up on the internet or anything outside of its training, which last happened in early 2022. Also, it doesn’t update its own knowledge as it chats with you.
- Content created by ChatGPT won’t pass AI detectors. Being in the business of analysing content, Google have researched this for years and established strong methods for detecting AI content. Also, inivisible watermarking, according to professor Scott Aaronson, guest researcher at OpenAI, “has the potential to become an industry standard.”.
Can I use it for SEO?
In some ways, yes. Google Webmaster Guidelines don’t prohibit tools like this from generating a meta description, for instance. This kind of use appears permissible, according to Google’s search relations guy, John Mueller.
You really shouldn’t use ChatGPT to generate and directly use keyword optimised copy. The temptation can be high for any marketing person to lighten their workload significantly by simply punching in a few instructions and being handed all the content they could publish. However, Google’s guidelines are centred around penalising auto-generated content that adds little value to the world, and as a result, doing this could put your website in a bad light and lose you revenue in the bigger picture.
Don’t get swept up in the hype that ChatGPT can perform miracles and write better than the great poets of history. While its capabilities are truly impressive, it is not a legitimate easy way of thinking for yourself. Sam Altman, the CEO of OpenAI said “ChatGPT is incredibly limited” and “it’s a mistake to be relying on it for anything important right now”.
I tried a personal experiment: There is a very obscure, hard to find trick in the Hubl programming language to bypass the fact that you can’t use a loop variable outside of said loop. I tested ChatGPT to see if it knew the secret. It first told me that Hubl loop variables COULD be used outside of the loop, like in other languages. When I challenged it, it apologised and admitted that these variables couldn’t be used. It wasn’t at all aware of the hidden method, nor did it even truly understand the question. It was just parroting information back to me, albeit in a very coherent and convincing (on the surface) way.
What can/should I use it for?
Since Google wants their results to be USEFUL to human beings, their guidelines and algorithms try to prevent and suppress spammy, automatically generated content in search. If however, a tool can generate content that is for all intent and purposes unique, accurate and truly USEFUL to human beings, then the line is more blurred as to what is considered legitimate.
The rule of thumb is to use the tech as a tool to help you better produce valuable content, faster and more effectively than you otherwise would have on your own, while avoiding adding spam to the world.
With ChatGPT, the fewer instructions you give it, the more likely it’ll write similar content to what it’s written before, or for other people’s requests. The more specific your request, the more specific, unique and specialised the end result.
For example: “Write me an essay” is going to produce a far more generic result than “write me an approximately 1000 word essay on the influence of the Gothic period on Edgar Allan Poe’s poetry. Write this in a fun and flippant style. Make sure to talk about the characters in Poe’s novels”.
Since the chatbot remembers what you said previously in a given conversation, you can reference things you mentioned earlier when refining your requests.
Just get ideas: Ask ChatGPT for a list of ideas for a summer vacation near your area and see what it says. Use the power of this tool as you see fit. It can be faster and less cluttered than a search engine.
Rather than using the chatbot to write your entire article, for instance, you can ask it to produce an outline of what you want, and use that to update your own knowledge on the subject. Remember that the accuracy is not always good, so double check every fact in Google and use your discretion.
You can use ChatGPT to explain a complex topic in a simple way:
Use ChatGPT to summarise information you already have into a small description. Examples:
- A meta description for a web page, which is only supposed to be about 160 characters long.
- A Twitter tweet, also limited to 280 characters in length
- A product description. If you have already written a long version with all the bells and whistles, ask ChatGPT to help you make a more succinct version.
Translate your content for an international audience. It’s advised that you get your work checked by a real language translator, but you could save time and money by getting help from ChatGPT.
Remember that these cases should be useful to a human being in the context that they are used. The objective is to help people find and process relevant information, not just to have your job done quickly so you can get more vacation time.
You can use it to perform tedious text formatting or editing tasks on long lists of data. For example, you can provide a long list of dates formatted as “first of January Twenty Twenty two”, and ask ChatGPT to format each date in the list to 1st Jan 2022.
You can write or debug computer programming code. When I started out as a web developer, I had never written a PHP web page before. My employer offered me a reward and asked me to build a template to use on a real client’s website. I agreed and wrote the page. It worked on my local computer, but when my employer tried to make it work on the client’s website, it broke things. I agreed to sit up late with him and debug the code manually. After hours of testing, we found a single full-stop that was out of place, and removing it fixed the entire page instantly. I’m glad to say I got the reward in the end. But had I had ChatGPT then, I could have saved myself and my employer a whole evening!
Have fun playing with the world’s latest powerful toy…
Stay In The Know
Cut the clutter and stay on top of important news like this. We handpick the single most noteworthy news of the week and send it directly to subscribers. Join the club to stay in the know…