The potential, limits and use cases of ChatGPT for automation
14 décembre, 2022
The potential, limits and use cases of ChatGPT for automation
👋 Hello everyone and welcome to our new subscribers,
We're delighted to welcome you to the twelfth edition of the SmartMyData newsletter. Every 15 days, we'll be bringing you the latest news from the world of automation and RPA.
This week, the #chatGPT is in the spotlight, with a focus on task automation.
Many software publishers are asking themselves how they can integrate ChatGPT into their applications. Smartmydata is doing just that, and will be looking into the matter once the cost and usage options have been defined.
Over the past few days, we've been hearing a lot about ChatGPT, the chatbot designed by @openAi (a San Francisco-based company whose founders include Elon Musk and Peter Thiel, as well as AWS and Infosys, whose aim is to promote and develop artificial intelligence solutions), the chatbot that has the answer to almost everything!
To introduce him to you, what better way than to ask him to do it himself:
"The chatGPT is a natural language processing (NLP) tool that uses state-of-the-art language modeling to generate human-like responses to user input. Developed by the OpenAI team, chatGPT is powered by the GPT-3 (Generative Pretrained Transformer 3) model, which has been trained on a massive amount of text data and can generate highly consistent and realistic responses to a wide range of topics.
One of the key features of chatGPT is its ability to understand and respond to context. This means that, unlike many other NLP tools, chatGPT can conduct a conversation in a way that feels natural and intuitive to the user. For example, if a user asks chatGPT a question on a specific topic, chatGPT will provide a relevant answer that takes into account the context of the conversation.
In addition to its conversational capabilities, chatGPT also offers a number of other useful features. For example, it can generate text in a variety of styles and formats, making it useful for tasks such as writing articles, generating reports or even creating social networking posts. It can also perform language translation, enabling users to communicate with chatGPT in their native language and receive responses in a different language.
A potential use case for chatGPT is customer service, where it could be used to provide quick and accurate answers to common customer questions. It could also be used in education, helping students to familiarize themselves with a wide range of subjects by providing personalized, engaging answers to their questions.
Overall, chatGPT is a powerful and versatile tool with many potential applications in a variety of fields. Its ability to understand and respond to context, as well as its ability to generate text in a variety of styles and formats, make it a valuable asset for anyone looking to enhance their natural language processing capabilities."
Here's a complete and impressive presentation of chatGPT. To demonstrate the power of this technology, we decided to carry out just one test and copy the first text proposed by the artificial intelligence. I think you'll agree with me when I say that the result is pretty convincing.
If you'd like to try it out for yourself, well :
Visit the dedicated page on OpenAI,
Log in with your e-mail address to access the interface,
Start the conversation with ChatGPT. Have a good chat !
❤️ Chatgpt business case
In the years to come, AI is set to bring significant changes to many sectors, including customer service, development, banking and insurance, accounting, office automation, translation, business intelligence, content creation... Here are just a few examples of what chatGPT AI is capable of doing in professional fields.
ChatGPT vs Development
ChatGPT lets you debug and correct lines of code with an explanation.
You can ask the ChatGPT if a piece of code contains a security flaw. The chatbot responds by explaining why the piece of code contains a flaw, and gives the correct code.
The chatbot generates lines of code automatically in a matter of seconds.
ChatGPT vs Contract Clauses
The chatGPT is capable of drafting clauses for a contract, as @Andrew Stokes shows us in his post. The chatbot drafted a payment clause thanks to indications provided by the general contractor who was processing :
- A main payment obligation
- Payment terms
- Payment disputes and suspension
- Late payment interest
- Termination for non-payment
- Costs arising from payment disputes
- Owner's performance bond to guarantee owner's obligations
ChatGPT vs. Digital content creation
ChatGPT generates well-structured posts and descriptions in less than 30 seconds. Jean-Baptiste Berthoux conducted a fun test pitting it against chatGPT, publishing a post each day written by the #IA or by himself. The result is quite interesting.
ChatGPT vs Product description
The cat is capable of generating human-like texts based on the information it is given and on which it has been trained. In our newsletter n°9 "Using AI for automation", we've also taken a look at automatic content writing, particularly for product descriptions, blogs, videos and quotes. Take a look if you're interested in automatic product description.
ChatGPT vs. app development
Twitter users asked ChatGPT to help them create an application, and it worked. As well as providing general development advice, it even provided sample code that could be used in specific scenarios. However, it's important to examine and check the code carefully before using it in your own project.
ChatGPT vs. Writing blog posts
The AI is capable of generating a blog post from A to Z in a matter of seconds. It can even generate outlines for your blog posts if you lack inspiration.
ChatGPT vs Information search
ChatGPT acts as a powerful search engine, enabling us to find answers to your questions quickly. Keep in mind that ChatGPT's AI is trained with data from events mainly prior to 2021. This means it knows nothing about current weather or other recent trends. Nevertheless, it can handle any kind of factual information very well!
ChatGPT vs Algorithm
An interesting use case led by @StéphaneSanchez to create an incoming e-mail algorithm.
Some additional resources :
- Transforming TypeScript code into Python
- My life as a developer is about to change
- Building an end-to-end MLOps architecture for an AI startup
- OpenAI's new ChatGPT bot: 10 coolest things you can do with it
For the moment, it's not clear how this will be used, or how much it will cost. If this isn't a limitation, it's still a big question mark. Data confidentiality is also a crucial issue. Are we on a Google-style model, or when it's free, are we the product?
From a business evolution point of view, the chatGPT phenomenon gives an image of artificial intelligence as a miraculous solution, having the answer to everything and putting many people out of work.
But is it real?
chatGPT is a solution on which we have very little experience. It's difficult to know what the risk of error is. But what's important to bear in mind is that AI often makes mistakes. In his latest post on chatGPT, @Benoit Raphael gives us several examples of recent mistakes made by the AI.
- The Stack Overflow site deleted ChatGPT's answers to questions posed by coders because the robot was sending back false information, which was becoming unmanageable for the company.
- On another AI, Bert the one from #Google, doesn't understand negation well which results in bad studies or summaries.
AI is well on the way to becoming infallible on many subjects, but let's not forget that it's still in the testing phase.
Here are the limits of ChatGPT as indicated by its creator OpenAI:
ChatGPT sometimes writes plausible but incorrect or absurd answers. Solving this problem is difficult, because :
- During RL training, there is currently no source of truth
- Training the model to be more cautious leads it to refuse questions it can answer correctly
- Supervised training misleads the model, as the ideal answer depends on what the model knows, rather than what the human demonstrator knows.
ChatGPT is sensitive to adjustments to the input formulation. For example, the model may pretend not to know the answer, but with a slight reformulation, may answer correctly.
Ideally, the model would ask clarifying questions when the user has provided an ambiguous query. Instead, current models generally guess what the user wanted.
The model makes efforts to refuse inappropriate requests, but will sometimes respond to harmful instructions or display biased behavior.
😆 Some amusing use cases created by AI
To round off this newsletter, we'd like to share with you a number of fun uses of chatGPT. The one that caught our eye the most was that of @YoanBernabeu. He submitted story pitches imagined by himself and his children to the AI, and the result was quite astonishing. He decided to create the site Histoires.io to bring together all the stories created by AI, accompanied by illustrations by #Dalle2 and #midjourney.
Also, a few other examples of recreational subjects
- Movie created via chatgpt
- Story created via chatgpt
And what stories will you create with chatGPT?