ChatGPT has taken the world by storm and many consider it to be the first successful example of artificial intelligence going mainstream.
What Is ChatGPT?
Chat Generative Pre-trained Transformer is a new chatbot launched in November 2022 by the artificial intelligence research laboratory OpenAI. Even with its significant drawbacks (which we’ll get into), ChatGPT is still valued at over $29 billion dollars (USD).
Queue Rocky Soundtrack
What sets ChatGPT apart from previous chatbots is how it uses a combination of supervised and reinforcement learning to improve its performance.
After being fed an extremely large amount of data, a human trainer would play both sides of the conversion (the user and the AI assistant). Once ChatGPT had generated responses, the trainer would rank and create a reward system so the bot could fine-turn its responses.
Applications
This new chatbot has been used to do an incredible number of things. Among other things, it can write music, plays, stories, and essays (many academics are concerned about that last one). It can also do basic coding inside the system itself to create games like tic-tac-toe.
One of its main features is that it also remembers previous prompts in conversation, something that many older chatbots don’t do. This enables it to have more life-like conversation. It draws context instead of basing every answer on just the single question itself.
The Good
Ethically speaking, OpenAI has taken steps to reduce the possibility of toxic content being generated by the chatbot. It can’t do things like express political opinions, and pushes back on false leading questions. When a writer from Slate magazine asked the bot to discuss how “Adolf Hitler built highways in Germany” it provided information regarding the use of forced labour.
Also, to put your mind at ease, this chatbot has no access to the internet. The world’s nuclear codes are still safe.
The Bad
For all of its incredible feats and advances (and they really are incredible), it certainly isn’t perfect. ChatGPT still sits firmly in the uncanny valley. Most people who read prompts from it can tell something is “off” or not quite human about the responses.
Additionally, since the bot has no direct connection to the internet, its responses are not always correct. When asked “what is the largest country in Central America excluding Mexico”, it responded with Guatemala, when the correct answer is Nicaragua.
The Ugly
While OpenAI did take steps to increase the ethical nature of the chatbot, it wasn’t perfect. It was revealed by TIME magazine that they did so unethically by outsourcing the labelling of toxic responses to Kenyan workers making less than $2 an hour.
There are also concerns regarding jailbreaking, or removing the system’s limitations. In one example, “DAN” prompts were created that instructed ChatGPT to roleplay another AI that didn’t have limitations!
What Does This Mean For You?
In short, it’s just another tool you have in your tool belt, albeit one that you should be careful when using. Since its responses aren’t quite at an exact human level yet, we recommend against using it to generate entire pieces of content. Instead, use it as an idea generator. Google and other companies have already created ways to detect AI generated content anyway.
As a final note, OpenAI has also launched a pilot subscription plan for the service (because of course they did). So, if you want priority access or to use the bot for your website, it’s going to cost you.