What is ChatGPT and how can you use it?
That’s what ChatGPT is and why it might be the most important tool since modern search engines.
OpenAI introduced a long-form question-answering AI called ChatGPT that answers complex questions conversationally.
It is a cutting-edge technology since it has been taught to understand what people mean when they ask questions. Many users fear its ability to provide human-quality answers, fueling the realization that it may eventually have the power to teach humans how to interact with computers and retrieve information.
What is ChatGPT?
Based on GPT-3.5, OpenAI created the big language model chatbot known as ChatGPT. It has the remarkable ability to communicate in the form of a conversational dialogue and provide responses that can appear surprisingly human.
Predicting the subsequent word in a string of words is a task carried out by large language models.
Reinforcement Learning with Human Feedback (RLHF) is an additional layer of training that uses human feedback to help ChatGPT learn the ability to follow instructions and generate satisfying responses to humans.
Who created ChatGPT?
The artificial intelligence company OpenAI, headquartered in San Francisco, developed ChatGPT. OpenAI Inc. Open is the not-for-profit parent company of AILP.
OpenAI is best known for its flagship DALL•E, a deep learning model that generates images from text instructions called prompts.
The CEO is Sam Altman, formerly president of Y Combinator.
Microsoft is a partner and investor to the tune of $1 billion Together, they created the Azure AI platform.
Major language models
ChatGPT is a large language model (LLM). Large language models (LLMs) are trained with large amounts of data to accurately predict which words occur in a sentence.
It was discovered that increasing the amount of data increases the ability of language models to do more work.
According to Stanford University:
“GPT-3 has 175 billion parameters and was trained on 570 gigabytes of text. For comparison, its predecessor, GPT-2, was 100 times smaller at 1.5 billion parameters.
This increase in scale radically changes the model’s behavior—GPT-3 is able to perform tasks it was not explicitly trained to perform, such as translating sentences from English to French, in which There are few examples of training.
This behavior was largely absent in GPT-2. Furthermore, for some tasks, GPT-3 outperforms models that were implicitly trained to solve those tasks, although it underperforms for other tasks.
LLMs predict a sentence and the next word in a series of words in the next sentence – like auto-completion, but on a mind-bending scale.
They are able to produce paragraphs and full pages of text thanks to this skill. But LLMs are limited in that they don’t always understand what a person wants.
And this is where ChatGPT improves on the state of the art, with the aforementioned Reinforcement Learning with Human Feedback (RLHF) training.
How was ChatGPT trained?
GPT-3.5 was trained on a large amount of data about code and information from the Internet, including sources such as Reddit discussions, to help ChatGPT learn and respond to dialogue in a human way.
ChatGPT was also trained using human feedback (a technique called Reinforcement Learning with Human Feedback) so that the AI learned what to expect when humans ask a question. This method of training LLM is groundbreaking because it goes beyond merely teaching LLM to predict the following word.
A March 2022 research paper titled Training Language Patterns to Follow Instructions with Human Expression explains why this is a breakthrough approach:
“This work is motivated by our goal to enhance the positive effects of large language models by training them to do what a specific set of humans want them to do.
By default, language models serve the purpose of predicting the next word, which is just a proxy for what we want these models to do.
Our results suggest that our techniques hold promise for making language models more helpful, accurate, and harmless.
Making language models larger does not inherently make them better at following user intent.
For example, large language models can produce results that are incorrect, toxic, or simply not helpful to the user.
In other words, these models are not connected to their users.
The engineers creating ChatGPT hired contractors (called labelers) to classify the outputs of two systems, the GPT-3 and the new InstructGPT (a “silencing model” of ChatGPT).
Based on the classification, the researchers reached the following conclusions:
“Labels significantly prefer the InstructGPT output over the GPT-3 outputs.
InstructGPT models show an improvement in accuracy compared to GPT-3.
InstructGPT shows a small improvement in toxicity compared to GPT-3, but not bias.
The research paper concluded that the results of InstructGPT were positive. Still, he also noted that there is room for improvement.
“Overall, our findings demonstrate that, while large language models’ safety and reliability are reduced, fine-tuning them using human preferences considerably enhances their performance on a variety of tasks. To improve, a lot of effort needs to be done.”
What sets ChatGPT apart from a simple chatbot is that it was specifically trained to understand human intent in a question and provide helpful, truthful and harmless answers.
Because of this training, ChatGPT can challenge certain questions and reject parts of the question that don’t make sense.
Another research paper from ChatGPT shows how they trained AI to predict what humans prefer.
The researchers found that the metrics used to rank the outputs of natural language processing AI resulted in machines that scored well on the metrics, but not as well as humans would expect.
Here’s how the researchers explained the problem:
“Numerous machine learning applications focus on improving straightforward metrics that are just approximations of designer intent. This may cause issues, such YouTube suggestions that promote clickbait.”
So the solution they developed was to create an AI that could output responses based on human preferences.
To do this, they trained the AI using datasets of human comparisons between different answers so that the machine became better at predicting what humans judged to be satisfactory answers.
The paper shares that the training was done by summarizing Reddit posts and was also tested on summarizing news stories.
Learning to Summarize from Human Feedback is the title of the research study from February 2022.
The researchers write:
“In this work, we show that it is possible to significantly improve abstract quality by training a model to optimize human preferences.
We collect a large, high-quality detaset of human comparisons between abstracts, train a model to predict human preferred abstracts, and use reinforcement learning using this model as a reward function. By doing so, the summary policy should be improved.”
What are the limitations of ChatGPT?
Restrictions on toxic reactions
ChatGPT is specifically programmed to not deliver toxic or harmful reactions. So it will avoid answering such questions.
The caliber of the instructions determines the caliber of the responses.
The fact that the output quality is largely dependent on the input quality is a significant ChatGPT restriction. In other words, expert prompts produce better responses.
The answers are not always correct
Another limitation is that because it is trained to provide answers that humans perceive as correct, the answers can deceive humans into believing that the output is correct.
Many users have discovered that ChatGPT can provide incorrect responses, including some that are wildly inaccurate.
Moderators at the coding Q&A website Stack Overflow may have discovered an unintended consequence of answers that feel correct to humans.
The stack overflow was flooded with user responses generated from ChatGPT that appeared to be valid, but there were many incorrect responses.
Thousands of responses overwhelmed the volunteer moderator team, prompting administrators to impose bans against any user posting responses generated from ChatGPT.
A flood of ChatGPT responses resulted in a post titled: Temporary Policy: ChatGPT is Banned:
“This is a temporary policy intended to reduce the inflow of responses and other content created with ChatGPT.
…the main problem is that while the responses generated by ChatGPT have a high error rate, they generally “look like” they “might” be good…”
The experience of stack overflow moderators with incorrect ChatGPT responses that appear correct is something that OpenAI, the makers of ChatGPT, are aware of and warn about in their announcement of the new technology.
OpenAI defines the limitations of ChatGPT.
The OpenAI announcement offered this warning:
“ChatGPT sometimes writes understandable but incorrect or nonsensical responses.
This problem is difficult to solve, as:
- There isn’t a source of truth throughout RL training right now.
(2) By training the model to be more careful, it rejects questions that it can answer correctly. And
- Supervised training misleads the model because the ideal response depends on what the model knows, rather than what the human demonstrator knows.
Is ChatGPT free to use?
Currently, ChatGPT is available for free use during “Research Preview.”
The chatbot is currently open for users to test answers and provide feedback so the AI can improve at answering questions and learn from its mistakes.
The official announcement states that OpenAI is eager to receive feedback on bugs:
“Although we have made an effort to block incorrect requests, the model occasionally obeys damaging commands or behaves biasedly.
Although we use the Moderation API to flag or remove certain categories of harmful content, we anticipate that some false negatives and positives may still occur.
In order to assist us in our continued efforts to improve this system, we are glad to gather user feedback.
There is currently a contest with a prize of $500 in ChatGPT Credits to encourage the public to rate answers.
“Users are encouraged to provide feedback on problematic model outputs through the UI as well as false positives/negatives from the external content filter that is also part of the interface.
We are particularly interested in feedback about harmful outcomes that can occur in real-world, non-adversarial situations, as well as feedback that helps us uncover and understand new risks and potential mitigations.
You may choose to enter the ChatGPT Feedback Contest3 for a chance to win up to $500 in API Credits.
Entries can be submitted through the feedback form linked in the ChatGPT interface.”
Will Language Models Replace Google Search?
Google itself has already built an AI chatbot called LaMDA. Google’s chatbot performance was so close to human conversation that a Google engineer claimed that LaMDA was sensitive.
Given how these large language models can answer so many questions, is it far-fetched that a company like OpenAI, Google, or Microsoft will one day replace traditional search with an AI chatbot?
On Twitter, some people have already predicted that ChatGPT will overtake Google.
The prospect that a question-and-answer chatbot could one day replace Google is frightening for those who make a living as search marketing professionals.
This has sparked debate in online search marketing communities, such as the popular Facebook SEOSignals Lab where someone asked if searches could move away from search engines to chatbots.
After experimenting with ChatGPT, I have to agree that the fear of search being replaced by a chatbot is not unfounded.
The technology still has a long way to go, but it’s possible to envision a hybrid search and chatbot future for search.
But the current implementation of ChatGPT seems like a tool that will require the purchase of credits at some point to use.
How can ChatGPT be used?
ChatGPT can write code, poems, songs, and even short stories in a specific writer’s style.
Mastery of the following guidelines takes ChatGPT from an information source to a tool that can be called upon to accomplish a task.
This makes it useful for writing essays on virtually any topic.
ChatGPT can serve as a tool for creating outlines for essays or even entire novels.
It will offer solutions for almost any task that has a written text solution.
Conclusion
As mentioned earlier, ChatGPT is envisioned as a tool that the public will eventually have to pay to use.
Within the first five days since ChatGPT was opened to the public, more than one million users had registered to use it.
Other links
I will do vector tracing and convert your logo or sketch into a vector