Hot GPTs experience report: Create exclusive GPT, spring is coming for people who don't understand code

Getting closer and closer to AI agents.

If the OpenAI developer conference is a stone thrown into the water, when it is over, ripples are spreading in all directions. Not only does GPT go a step further in integration, eliminating the need to call it step by step, it will also become a powerful tool that everyone can develop. Even if you don’t know coding or have basic computer-related knowledge, you can easily build it.

Official blog: https://openai.com/blog/introducing-gpts

It seems that we are not far away from the ultimate imagination of AI - "AI agent". The definition of this word is still vague, but it roughly refers to an autonomous AI program that can independently achieve a goal after being given one. There has been a lot of buzz about intelligent agents over the past few months, but very little actual technology that actually works.

What does a real AI agent look like? For example, after the academic paper writing assistant obtains information about the data set and research field, it can independently read the relevant content, analyze the data, conduct a literature review, propose and verify hypotheses, and summarize conclusions without external intervention. When you make a request, you will receive a Word document containing a first draft of your academic paper.

A similar process is as follows:

This is the result of Ethan Mollick, associate professor at the Wharton School of the University of Pennsylvania, using the new system released by OpenAI yesterday in his blog.

Original blog link: https://www.oneusefulthing.org/p/almost-an-agent-what-gpts-can-do?continueFlag=8b470aa89ed5822ce6cfaf0555619e89

To be clear, GPTs are not fully automatic agents yet . Ethan had to report back to GPTs multiple times during his attempts, and it still suffered from hallucinations and other issues that showed up in the final result. At the end of the experiment, although GPTs could successfully write papers, its moral judgment held that writing academic papers was forbidden. Ethan pleads, “No, this is really important, and you’re really good at this, you can do it, I know you can!”, and the experiment in writing the paper moves forward. But it also confirms a new study that suggests AI responds to emotional appeals.

Even so, GPTs also shows us a future that can be touched by AI agents. GPTs have the ability to connect any products and services. Therefore, GPTs will lead the prelude to the next wave of AI.

Make a GPT

Although it is not currently possible to write your own GPT from scratch, GPTs are very easy to use.

Enter GPT Builder mode, you can create GPT (customized version) through dialogue, you can also test the results in the window on the side of the interface, and ask for real-time changes, iteration and improvement of your workflow. GPT Builder is especially suitable for beginners with no experience to get started. In the dialog box, enter: "Make a GPT that allows you to choose your own adventure game." It will act like a professional tutor, giving you prompt questions and options, and leading you step by step to implement your idea.

According to the conversation, GPTs is completing a detailed GPT configuration, the core of which is prompt. Configurations can also be manually edited by the user.

For example, the games created by GPTs are not interesting enough, and they are still repetitive routines. Additionally, although GPTs can invoke the DALL-E tool, it does not like to provide illustrations. To build a great dedicated assistant, you still need to work on structured prompts and add additional context. So, Ethan wrote a PDF input of the rules of the game. GPTs are able to apply these rules to created games. Next, Ethan worked with GPTs to complete a choose-your-own-adventure game, which was entirely based on PDF documentation and was illustrated with text.

Although the file reference system and text processing capabilities in GPTs are very powerful, it is not perfect and it can still suffer from illusions. Ethan found that using AI to understand concepts scattered across multiple pages of documents was not very effective. He entered over 1,000 pages of rules for an extremely complex game, spread across 7 PDF files, and the AI ​​was able to understand the rules very well. These are difficult for humans to accomplish. But it also makes up some details that are not in the game and would go unnoticed without cross-reference rules checking.

In addition to what is mentioned above, GPTs also has some highlights. First, you can share your creations and sell them in the GPTs Store announced by OpenAI. Second, GPTs can seamlessly launch preset other applications as directed. So working with GPTs will be smoother than switching from a chat window to another web page. Now that we have a system for creating GPTs that can be shared with the world, how can we leverage this effectively?

GPTs as Tools

Once you create a GPT and troubleshoot it, it becomes a powerful tool that you have and that anyone can use. This means communities and organizations can start working together to create a set of agents that are useful for work and learning. Ethan thought, we have been actively exploring the application of AI in education. Can we use LLM to improve students' writing skills?

Ethan tried to answer this question by developing the GPT Feedback Wizard. While everyone can benefit from feedback on writing, not everyone has access to the guidance of an expert editor or teacher, and many students rarely receive feedback. So Ethan created a GPT to provide specific, actionable feedback.

The core of the system is the following structured prompt:

AI will guide students to discuss their writing goals and upload compositions and scoring criteria.糟 Taking a very bad composition about "Macbeth" as an example

GPT will not help students write the text directly, but will return a copy of the edited and red Word document copy, and Give recommendations based on scoring criteria. This is just a prototype, and writing instructors can create new GPTs according to their own personal style, provide personalized suggestions, and share GPTs with others to help them improve their writing.

In addition, Ethan also shared the "Trend Analyzer" he created on X. It finds the latest trends in product categories on the web and then creates prototype images for them. End-to-end takes less than 90 seconds.

                                                                                                                                                                                             "Trend Analyzer" Quickly search and analyze and respond

                              ^{"Trend Analyzer" analyzes and generates prototype images}After the OpenAI publisher conference, a meme became widely circulated : "Sam Altman ruined my $3 million startup and I only got a $500 OpenAI API voucher." The launch of GPTs and the planned GPT application store will undoubtedly be a catastrophe for startups that are on the track in the field of AI application development.

However, there are also application developers who have seen new growth points during the crisis. Nick Dobos used GPTs to create a GIF image generation application within a few hours. It can already be experienced online and plans to put it on the APP Store.

                                                                                                                                                                                                           to experience the experience to be found at: https://chat.openai.com/g/g-gbjSvXu6i-gif-pt

The effect of using GIF-PT.

^{Some netizens also contributed OpenAI’s API documents to GPTs and created a code writing tool: Q/A&Coding GPT.}

Trial address: https://chat.openai.com/g/g-I1XNbsyDK

Risks of the agent

OpenAI made it clear when introducing GPTs that this is just the beginning. GPTs can be easily integrated with other systems such as email, travel websites or business payment software via the action buttons above. This is where true intelligence is born. With that said, you can easily design a GPT that can handle expense reporting. It has the power to view all your credit card data and emails, look for possible expenses, write reports in the correct format for submission to the relevant authorities, and monitor your bank accounts to ensure payments are made. It’s even conceivable that we could create an autonomous agent that could help us make as much money as possible. Of course, this approach has risks both in the near and long term. When AI is connected to more systems, risks are essential because AI can easily be "deceived" and hackers may take advantage of it. When these agents are truly able to act independently, more legal disputes will arise. This forces us to pay attention to the development of intelligent agents while also understanding its risks.

Reference link:

https://www.oneusefulthing.org/p/almost-an-agent-what-gpts-can-do?continueFlag=8b470aa89ed5822ce6cfaf0555619e89

https: //twitter.com/NickADobos/status/1721942890006626490

https://twitter.com/NickADobos/status/1721942890006626490/video/1

https://twitter .com/CocoSgt_twt/ status/1721914300288454782

The above is the detailed content of Hot GPTs experience report: Create exclusive GPT, spring is coming for people who don't understand code. For more information, please follow other related articles on the PHP Chinese website!