The Future’s Bright, But Ethical Waters Murky


The Gist

  • Tech triumph. OpenAI introduces GPTBot to supercharge AI systems.
  • Ethical enigma. Debates rise over data usage and copyright concerns.
  • Web wonder. Potential to expand ChatGPT knowledge past September 2021.

The ever innovative minds at OpenAI have just unveiled GPTBot, a web crawler that could give a significant boost to the performance of future AI models, including GPT-4 and the much-anticipated GPT-5.

ChatGPT has some work to do in order to move past its current knowledge cut-off of September 2021. If you think of the Internet as a vast library, GPTBot is there to scour the collection, looking for pieces of information to pick up in order to help AI systems become better with more accurate, relevant and timely responses.

This latest news comes on the heels of OpenAI’s debut of enhanced conversational capabilities, including prompt examples and suggested replies, just last week.

Related Article: What’s Behind ChatGPT’s Latest User Experience Update?

How Does OpenAI’s GPTBot Work?

Basically, GPTBot is on a mission to find information on the internet that can make AI systems smarter, more capable and safer.

The data it collects can help these AI systems in several ways.

  • Accuracy. The data can help the AI systems give more correct answers or predictions.
  • Capabilities. The data can help the AI systems learn to do more things.
  • Safety. The data can help the AI systems understand better how to avoid doing things that could cause problems or harm.

Related Article: Can Tools Like ChatGPT Help Personalize Marketing Strategies?

Is There a Downside to OpenAI’s New GPTBot?

It appears that OpenAI is inching its way back into a better connection with the Web after pulling its browser plugin over concerns related to paywall access. With GPTBot, the aim is to ensure that ChatGPT doesn’t provide information that’s restricted behind paywalls. OpenAI claims GPTBot is very careful about which web pages it looks at, avoiding any sites that require payment to access (paywalls), as well as sites that collect personal information about people (like names, addresses or phone numbers). It is also trained to sidestep sites with content that bump up against OpenAI’s rules.

But this latest announcement from OpenAI has caused some debate within web forums about whether it’s ethical or legal to use information from the web to train AI systems that can make money — some think if OpenAI is making money from this, they should share the profits. Others worry that copyrighted material (like text, images, videos, music) could be used without providing proper credit.

On the other hand, many think it’s fine for OpenAI to use public information from the web — just like anyone else can.

How to Opt-in to OpenAI’s GPTBot

Web crawlers typically scan and index publicly available web pages across the internet and unless you opt-out, GPTBot could access your site automatically. However, if a website owner only wants GPTBot to look at certain parts of their site, they can specify which parts are okay and which are not, like this.

User-agent: GPTBot

Allow: /directory-1/

Disallow: /directory-2/


Source link