A foreign boy spent his own money to help the AI ​​learn to order pizza online. Netizens wondered: Will the AI ​​be hungry?

Imagine that on a sunny afternoon, an AI has just finished outputting a lot of requests from users, and opens the takeout APP out of boredom, planning to order something.

Wait, AI? Why does AI order takeout?

Don’t worry about why you order takeout, the important thing is that you can order it.

Today I want to introduce such a startup company-HyperWrite.

Since ChatGPT became popular, companies engaged in various types of AI have sprung up like mushrooms after a rain.

Recently, HyperWrite has released such an AI agent that is still in the experimental stage. This agent can browse web pages at will like a human, and can even interact with websites.

In the just-concluded demonstration, HyperWrite CEO Matt Shumer showed how the agent accessed the Domino's Pizza website through a Chrome extension.

Even, this AI can find the address on its own, fill in the zip code, and pay to complete the order.

But then again, will the AI ​​be hungry?

This AI Can Order Pizza

Although the demo ended before the user entered their credit card number, Mr. Schumer It said that this feature is currently only available to some test users. The model functions as a personal assistant to the user, automatically handling basic network tasks based on instructions.

If you follow the artificial intelligence sector on Twitter, you can see the amazing potential of AI.

The AI ​​industry is filled with AutoGPT and BabyAGI. These AIs, driven by other models such as OpenAI’s GPT-4, are able to automatically generate instructions, further develop and execute them based on initial input, and further generate new instructions.

Currently, applications built on the GPT API are basically limited to single use, which means that the model cannot obtain information from previous interactions.

Schumer said that HyperWrite’s personal assistant AI is the first user-friendly AI that can operate a browser like a human.

However, experts remind that this type of AI may also have various vulnerabilities, such as phishing, fraud, etc.

Safety issues cannot be ignored

Schumer said, "Our team has realized the importance of AI being able to complete various tasks autonomously. However, this ability may It can be abused to hack into websites, manipulate communications, etc., and may even spread spam on social platforms."

Schumer said his team is very concerned about security issues.

They want to find the right way to deploy HyperWrite while ensuring safety for the entire society.

For me personally, since HyperWrite can order pizza by itself, will I order something else out of control tomorrow?

There is a precedent for this kind of thing. Therefore, Schumer's team is committed to ensuring its security based on the application of HyperWrite.

He explained that proxies like AutoGPT and BabyAGI might be thought of as simple tools that accept text input and provide output.

Users can integrate them into a system that can adapt to new information and improve it over time. Basically this process can be cyclical until it reaches up to the expected goal.

Although GPT is powerful, it is not so advanced that everyone thinks it is dangerous.

At present, the application of GPT in real life is still somewhat limited.

While it can perform basic research tasks, Schumer's team is focused on using GPT to improve business productivity.

The above is the detailed content of A foreign boy spent his own money to help the AI ​​learn to order pizza online. Netizens wondered: Will the AI ​​be hungry?. For more information, please follow other related articles on the PHP Chinese website!