[ad_1]
In recent weeks, the ChatGPT hype has blown up my tech-heavy social network. I follow a lot of coders and content creators on TikTok and Twitter, and most of them are Loss of consciousness Because of the disruption that OpenAI’s ChatGPT represents for their disciplines. Their common refrain: “This changes everything.” After trying it myself (more on that later) and reading more about it, I think I share their view.
As someone who follows the topics of technology and artificial intelligence, I should not have been surprised to see this review of the ChatGPT project. For many months my news feed has been full of threads about AI tools like GPT-3, GitHub Copilot, DALL-E, Stable Diffusion and more. But until now, these technologies required a certain amount of skill and hardware configuration that limited them to the realm of the technologist or savvy enthusiast. ChatGPT makes this technology available to anyone who wants to create a free user account on the service. This unprecedented ease of access has caught everyone’s imagination, including mine, and I admit I didn’t see it coming so soon.
Explaining the inexplicable AI
In some ways, the popularity of ChatGPT makes it easier for me to tell my friends and family what we do at SAS. As an artificial intelligence company, we commercialize the application of machine algorithms to business challenges and world-changing initiatives. When I tell this to my wife’s Aunt Susan, she shakes her head and says, “Ohhh,” even as her eyes glaze over. But when I say, “For example, we use natural language processing and reinforcement learning to build expert systems that can improve outcomes—you know, like in ChatGPT,” suddenly we have a common understanding (up to a point).
People I talk to feel one of the following about ChatGPT:
- Excited and optimistic: This includes students and new coders looking forward to using a tool like this to tackle their “mundane” tasks like writing essays, programming, ideation, and more. These people are not going to “cheat with AI”, but they see it as a tool to start their projects. This is better than opening a blank document or code file and starting from scratch.
- Scared and intimidated: This includes professional writers and content creators as well as veteran programmers. They acknowledge that ChatGPT’s results are often inaccurate or incomplete and potentially infringe the intellectual property of other creators, and in no way trace or cite sources. While ChatGPT’s failures can be comforting (“It can’t do my job yet!”), the fear is that many users will accept the results as “good enough” and thereby ruin the craft we’ve dedicated our careers to. .
GPT-3—the large language model behind ChatGPT—was trained on a variety of published materials, including books (fiction and nonfiction), web pages, social media, and scholarly journals. The model incorporates an astronomical number of paths, allowing it to predict a good response from almost any query. However, he is unable to cite sources or explain how he arrived at this answer. This makes the model vulnerable to (inadvertently) plagiarizing another source or (of course) presenting an answer that is incorrect or biased by some hidden parameter.
Think about it this way: When you use a calculator or computer program code to calculate “2 + 2,” the machine performs arithmetic By adjusting the bits in the memory register to get the correct sum of 4. prophesies – With a high degree of confidence – that the answer is 4. This way, it works more like your brain, which can come up with quick answers that are the result of all the training and reinforcement that life has given you. It’s only when you’re asked to “show your work” that you actually do the math and show your skill.
What does ChatGPT mean for self-service support?
Of course, one of the first things I did with my new ChatGPT account was ask it to solve some SAS questions. My team manages the SAS Support Communities, the largest repository of peer-to-peer knowledge among SAS practitioners. I want to know if this AI is fighting for my cause.
As far as I know, the content of our community site was used to prepare the GPT-3 model! Our friends at StackOverflow, the world’s largest general-purpose programming question-and-answer site, have taken the extraordinary step of banning GPT-generated answers from their forums. We haven’t done this on the SAS Community because we don’t yet see people providing AI-generated answers. (But that begs the question…how can we tell?)
The first question I tried was the most popular SAS topic we see for beginners: “How do I convert a character value to a number?”
ChatGPT has done a good job here. It explained the INPUT function and the role of information in reading a value as a number. Others may find some fault with it, but I found this answer acceptable to help beginners.
My next topic, a simulated homework assignment, yielded mixed results: “Write a SAS program to generate a data set of 100 random integers, each with a value between 1 and 100.”
I was impressed that the modeler wrote a program that uses the RAND() function with the “integer” method, since this is a relatively recent technique introduced in SAS 9.4. Unfortunately, the model got the wrong syntax, so the code won’t work as is – the RAND() function needs more arguments. Wow! That still leaves something for the student to find out for himself.
For my last example, I asked a question that contained some jargon: “Write a SAS program that shows how to implement LOCF.”
Nice to explain LOCF! “LOCF (last observation carried forward) is a method of imputing missing data by replacing missing values in a time series with the last non-missing value.“And many SAS coders try the LAG function, as ChatGPT did here, as a way to achieve that goal. Unfortunately, like many people, the model fell into the trap of thinking that the LAG function just ‘looks’ at the previous record.”
These mistakes are understandable… but then the model got cocky and offered more details than I asked for. And it became more wrong.
This code demonstrates many of LAG’s functions and attempts to apply the technique to multiple variables by using the OF keyword to specify a range of variables. I understand all the words that “come out of his mouth”, but they make no sense here. It reminds me of a talkative person at a party who is trying a little too hard to show you what they know, but you know they should have left before they were ahead.
We are all training the algorithm
I’m not trying to criticize ChatGPT by picking his answers to technical questions. The ChatGPT team intentionally allowed the model to generate answers that might be wrong. If they had smothered them with more certainty, then he wouldn’t have tried to answer nearly as many questions. Despite its flaws, I think it’s an amazing step forward. I also know that it only gets better.
Using ChatGPT (as millions have), we all work to train the algorithm. It relies on Reinforcement Learning for Human Feedback (RLHF), which means it learns when we rate its responses (thumbs up or down). And rumor has it that the next generation of the model, GPT-4, will be trained on much more data and will be released soon.
More chat about ChatGPT
If you want to learn more about GPT-3 and ChatGPT, I recommend this podcast from the Hard Fork team.
And here are some deep learnings from OpenAI: Predicting the potential misuse of language models for disinformation campaigns – and how to mitigate the risk. Obviously, they know that this technology has great potential, but also poses some risks.
Some of our SAS users have had fun with the tool: Rhyming verse SAS vs. Rs. And here’s one Creates a graph using SAS and then Python. (I’m proud to say that the SAS version was more concise and easy to read…)
You can also download this eBook on Natural Language Processing.
[ad_2]
Source link
