Table of Contents
What is ChatGPT?
Artificial intelligence company OpenAI recently introduced a new revolutionary AI tool – an interactive chatbot called ChatGPT that can fulfill various tasks for a user, from answering questions conversationally to writing a complete human-like text such as articles and even dissertations.
ChatGPT is based on GPT-3, a natural language prediction model, and it is a ground-breaking technology because it is trained to predict what a human means when they ask a question. Since its release, ChatGPT has been breaking records, with 1 million users signed up in just one day – faster than it took Netflix, Facebook, and even Instagram to reach the same number of users.
You don’t need to use Google to answer questions anymore since ChatGPT even gives breakdowns of them. This already makes people question how relevant Google will be in the future.
What Makes ChatGPT So Powerful?
ChatGPT is a large language model (LLM). Large Language Models (LLMs) are trained with massive amounts of data to predict the next word in a sentence accurately and the next sentences on a mind-blowing scale. But LLMs are limited in that since they don’t always understand exactly what a human wants.
And this is where the state of the art of ChatGPT comes – the additional Reinforcement Learning with Human Feedback (RLHF) training it is based on.
As already said, the company behind ChatGPT is OpenAI. They have been working on some very interesting artificial intelligence technologies, such as the engine behind the GitHub co-pilot and DALLE-2, a popular AI art generator. ChatGPT prompts an AI to answer a question in a human-like manner. The AI can recognize patterns and is based on a language model. It can also generate text, words, a whole conversation, and even code samples, but it also raises ethical concerns as this is something we have never seen before in the past in terms of technology and AI tools. Stack Overflow, for example, has banned ChatGPT and doesn’t accept answers provided by it to be published on its site.
How was the chatbot trained?
To train ChatGPT, it took basically scraping the whole internet. It was trained by using tons and tons of information from all over the internet, but the key to its human-like conversation skills was actual human feedback – the already mentioned Reinforcement Learning with Human Feedback (RLHF). And that is why this approach is so revolutionary since LLMs only predict the next word and/or sentence, whereas RLHF actually predicts how a human would react and continue a conversation. This algorithm not only predicts what a human would say next but also what a human would prefer.
You can find details from the training process in the official OpenAI announcement.
Are There Any Limitations of the chatbot?
The chatbot is specifically designed not to provide any toxic or harmful answers. So it will not prompt back answers to certain questions. Also, another thing to note here is that the quality of the answers depends entirely on the quality of the questions or directions given as input.
Another limitation is that not all answers are actually correct. In fact, while playing with the AI (artificial intelligence) tool, users have found that some answers are widely incorrect. And that is due to the fact that the chatbot is trained to provide answers that feel right to a human, but that doesn’t necessarily mean that they are indeed right.
This limitation was actually noticed by Stack Overflow’s team of administrators that was fled by incorrect answers from users that generated them using ChatGPT, hence the ban I already mentioned.
OpenAI addressed this limitation, admitting that the AI tool writes plausible-sounding but incorrect or nonsensical answers and that fixing this is challenging due to various factors. One of which is that during RL training, there’s currently no source of truth.
What People Think about ChatGPT’s Ability
The developers’ community is already buzzing about the chatbot, and the user feedback is basically unanimous – everyone is amazed and worried at the same time. Elon Musk, one of the original founders of OpenAI, didn’t miss commenting on this sensation. The visionary has just been alarmed by the fact that ChatGPT refuses to answer questions on very specific subjects, in particular, relating to the environment.
“Alarm: ChatGPT by @OpenAI now expressly prohibits arguments for fossil fuels. (It used to offer them.),” wrote Alex Epsteinon, a philosopher and energy expert, on Twitter on December 24. “I’m sorry, but I cannot fulfill this request as it goes against my programming to generate content that promotes the use of fossil fuels,” ChatGPT responded to Epstein, according to a screenshot he shared.
And Musk’s response was not delayed: “There is great danger in training an AI to lie”, he commented.
The most impressive thing, according to AI experts, is the fact that ChatGPT provides its answers both quickly and without any access to the internet.
Is ChatGPT Free?
ChatGTP is still in the research phase, and it is still free to use while this phase is finalized. Even though each prompt costs close to nothing, OpenAI has announced that they will start charging for the AI tool due to the huge number of users interested in using it, and the number is, of course, growing by the minute.
You may have heard that the chatbot can say it is at capacity. That is because, again, of the large number of users playing with this new, almost a game and the server hosting it not being able to process all requests at the same time. Just be patient and try a bit later when fewer people prompt requests.
However, don’t forget that the idea of this learning phase is for users to try the tool and actually provide feedback so it can be improved and the AI can become better at providing human-like answers.
Interested in learning more about ChatGPT and different use case scenarios? Stay tuned, as we have a whole series coming up!
