(Comments)
Last week, OpenAI announced the release of an AI language model called ChatGPT . ChatGPT is a chatbot based on the GPT-3.5 model, which in turn is an improved GPT-3 model, known from e.g. writing a scientific paper about myself , writing journalistic articles or writing poetry and prose .
ADVERTISING
What is ChatGPT? A model that conquers the internet
The ChatGPT model was trained using the Reinforcement Learning from Human Feedback (RLHF) method. This training was based on conversations with human trainers who impersonated both AI users and assistants. The trainers had access to the suggestions generated by ChatGPT, which were corrected by them and reported back to the AI as correct.
In turn, to create the reward model - essential for reinforcement learning - OpenAI engineers collected comparative data consisting of at least two model responses ranked by quality. To collect this data, OpenAI researchers used records of conversations between AI trainers and a chatbot, then randomly selected messages written by the model, matched them with alternative endings and had the trainers evaluate them in terms of content and fact.
How does ChatGPT work?
For the ChatGPT OpenAI training, over 45 terabytes of training data were used, which consisted of Common Crawl content , various online texts, books, and Wikipedia content . It is worth noting that the ChatGPT training was completed at the beginning of this year, which means that the model is not aware of, for example, the takeover of Twitter by Elon Musk or the death of Queen Elizabeth II.
As OpenAI declares, in addition to generating text based on the instructions we will introduce and talking in natural language, ChatGPT is also able to recognize errors in the codes of various programs and correct them. Created by OpenAI, ChatGPT refuses to answer questions about illegal activities or socially considered sensitive and controversial topics.
OpenAI also mentions the limitations of ChatGPT, noting that "ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers." According to OpenAI, this problem is due to the fact that training the model to be more careful led to it rejecting questions it cannot answer. On the other hand, supervised training misleads the model because the ideal answer depends on what the model knows, not what the human interlocutor knows.
In an ideal ChatGPT scenario where the user makes an ambiguous query, the model should ask him clarifying questions. Instead, the ChatGPT model guesses what the user meant.
What can ChatGPT write? Beautiful language, but often nonsense content
Despite its limitations, ChatGPT gained immense popularity on the Internet, as OpenAI made the model available for free to all interested parties for the time of open research. While it's still free and unlimited, I decided to talk to ChatGPT.
Unlike many intelligent models that conquered the Internet, ChatGPT also understands Polish, and this gave a fairly wide field to check whether it was possible to get something more out of it than generating advertising texts. Spoiler: yes.
After registering on the OpenAI website and logging in, I immediately started the conversation. The first task: "the content of Christmas wishes for parents". A relatively simple query, which additionally has quite a large amount of training data, was not a problem for ChatGPT.
It was similar to generating a soy candle ad on Instagram. It's probably not the most engaging text and many marketers could do better, but it presents the product in natural language and in a positive, buying light.
Then I threw ChatGPT into deeper water and asked why people prefer iPhone over Samsung. Artificial intelligence gave me the reasons that most often appear in the content of comparing Apple equipment with other brands.
However, when asked about polynomials, the SI started choking and did not finish the statement.
Then I threw ChatGPT into even deeper water, commissioning to write simple texts that every student of primary school, high school or technical school will collide with sooner or later. If there are smarties who are convinced that ChatGPT will do them at the last minute, then I have to cool down the expectations. SI writes total nonsense that would lead to failing the final exams.
A rather interesting example is ChatGPT's commission to write a sermon based on the letter to the Romans, specifically verse 21 in chapter 12.
Do not be overcome by evil, but overcome evil with good.
Rz 12,21
ChatGPT, instead of referring to the quote, paraphrased passages from other books, while not finishing his sermon.
ChatGPT has its uses, but it won't replace a human yet
ADVERTISING
Although some of the content generated by ChatGPT is not true and can at best bring a smile to your face, it is worth distinguishing factual errors and factual errors here. Although ChatGPT can, colloquially speaking, write nonsense and make cardinal factual errors with more difficult issues, each of the messages I received from the bot was linguistically, grammatically and stylistically correct.
Although reading that Seweryn Baryka is the main character of "Lalka" by Prus, I cannot hide my amused expression, I cannot deny that the model is extremely efficient - especially considering the realities of the Polish language, which does not have such a large amount of data training languages, such as English or Spanish.
Share on Twitter Share on FacebookWhy being the minimalist is good for your money
read moreWhat will ChatGPT answer if thre is a question? Write a story which relates the events of your life from past, present and future to the life cycle of a star. Use also those elements which could symbolize something or anything in your story.
read moreChatGPT is trained on vast amounts of online data in order to generate responses to user prompts. While it has gained traction among users, it has also raised some concerns about inaccuracies and its potential to perpetuate biases and spread misinformation.
ChatGPT passes exams from law and business schools
Linear regression is a method we can use to understand the relationship between one or more explanatory variables and a response variable.
While some celebrities enjoy fame and fortune, others prefer to take the cash and then return to their normal lives, or pursue a different dream. Whether this departure from the limelight is temporary or for good, these former A-listers are now working day jobs that more closely resemble our own.These Former Stars Quit Hollywood for Regular Day Jobs
Stacey Dash ©Eugene Powers/Shutterstock.com
How do I remove a specific value from an array? Something like:
read moreThe series was inspired by mythology and Viking sagas and is set 100 years after the events of the original Amazon Prime Video series.
Is there a reason why there are two different commands to generate a new variable?
Collaboratively administrate empowered markets via plug-and-play networks. Dynamically procrastinate B2C users after installed base benefits. Dramatically visualize customer directed convergence without
Comments