I was excited to use browsing but it has been pretty unusable for me so far. More often than not, it gets stuck in a loop or just displays a bunch of “Reading content failed”. Even worse, it has been hallucinating a lot. For example (as you can see in the screenshot below), it was looking up a URL and website I mentioned nothing about and this has been a common recurrence, in addition to it getting stuck. Bottom line, browsing has been a very poor experience for me and I’ve just gone back to using GPT-4 Default.
I feel like OpenAI rushed to release this feature to the Plus masses without longer testing to counteract Google’s recent waves and momentum in AI news. Anybody else having a similar experience with the browsing plugin?
3 Likes
I’d focus on identifying where it works best and learning how best to use it there.
It’s still a “Beta” feature so everything you’ve mentioned is probably to be expected. Although having GPT-4 connected to the internet to do actions is going to be incredible long term so having first-user advantage is a v good thing.
3 Likes
I think it’s great that they are working on it and letting us try it. However, it is very wonky right now, yesterday I got a bunch of “click failed” errors on websites that don’t have relevant robots.txt restrictions.
Looking forward to the feature maturing further.
For Internet access, I don’t know why anyone wouldn’t be using mighty Bing, this would allow the folks to mature the service for GPT-4 and at the same time, get’s GPT-4 prowess in form of the new Bing.
Bing right now is literally operating at different level and looks to be customer’s ready and still running the lovely GPT-4 model.
I can’t even use it to access OpenAI’s own website today! Not sure if it is workload issue or something else. Overall I think this feature has some room for improvement.
1 Like
I believe the main reason is that Browsing uses its own User-Agent instead of BingBot, which may lead to being blocked by websites when accessing a lot of content. Additionally, Browsing has overly strict “morals,” which means that if a website’s robots.txt file doesn’t specifically allow access for the ChatGPT-user user agent, it will result in a “click failed” response.
Overall, I don’t think this is a technical issue with OpenAI’s technology. Nowadays, web scraping, even for rendering Single-Page Applications (SPAs), doesn’t pose significant technical challenges. The main issue lies in Browsing’s own moral limitations.
for more information, check at: OpenAI API
2 Likes
I have experienced the problems you mentioned at times, but it has also been useful as well. It doesn’t always get stuck or caught in a loop. I asked ChatGPT about these problems previously and I got a more comprehensive answer, but I couldn’t find it in my history. Before it mentioned that some sites have robots.txt and other restrictions that prevent it from browsing them. Here’s what it replied when I asked it just now.
1 Like
Well, I finally found a guy explaining the right solution on YouTube which worked for me for this “Reading content failed” error. I hope it will work for all the people having this problem. Here it is: How To Fix "Reading Content Failed" Error in ChatGPT in ONE SIMPLE STEP - YouTube
