I am interested in learning more details about the browsing capabilities available when creating a custom GPT model. Specifically, when the ‘browsing’ option is enabled on the Builders page, is the GPT able to freely browse the entire open internet via search engines like Bing? Or is the browsing limited to only querying certain whitelisted APIs, databases, and approved sites?
My goal is to better understand what content a custom GPT could potentially access so I can avoid attempting to build models that require unavailable functionality.
I would appreciate any clarity you can provide on how the browsing works technically, including details about any whitelisting of sites/APIs, filtering, or content moderation that is applied. Do custom GPTs have unfettered internet access similar to a human user? Or does OpenAI restrict browsing to prevent risks around data privacy, scraping content, etc? Any insights you can share would be very helpful for my understanding. Thank you!
So there is a whitelist for specific gathering sources in specific areas. I can’t prove it as I’m on mobile but it can be found in local storage. If I can recall correctly it was mainly people search engines and news sources.
There is also an “opt-out” feature that the CHAT GOT crawler respects via robots.txt, so it may be that either
A) It will say it can’t
B) It will choose a different source
It also does tend to lean on avoiding privacy and copyright material. There was an interesting article that captured recent updates. I’ll see if I can find it.
You should just try it out yourself though
I’m very interested in learning more details about how browsing capabilities work technically for custom GPTs. However, I’ve had difficulty finding in-depth official documentation explaining the functionality.
Does anyone have technical insights to share on whether browsing is unfiltered or limited to whitelisted sources? I’d be grateful for any specific information or documentation sources that provide clarity on what custom GPTs can access through browsing and how it is restricted compared to human internet browsing.
I recognize developing thorough documentation is challenging, but it seems like an area that could be aided by OpenAI’s own models. Perhaps an idea for the future is leveraging GPT to automatically generate documentation covering frequently asked questions about technical aspects of the products? Even brief technical primers would be hugely valuable for developers exploring applications. I’d love to hear others’ thoughts on improving the availability of technical documentation and insights.