Currently using context URL in production to read specific pages and extract data. Previously (>3 weeks ago) the tool was functioning as intended and returned accurate data.
However, I noticed in the past couple of days the response from the model is often empty (there is no response) in both studio and via the api call. Additionally, sometimes it will say it accessed the url but the data will be incorrect and it does not cite the source (even in studio when it always used to).
Has anybody else noticed these issues recently? It is a known bug/issue? I realize this is a somewhat finicky feature but seems to be occurring more just recently.
edit: it seems google may use the last index of that specific url. For real time pages, it seems to be out of date (sometimes 1-2 days old). is this correct?
Hello,
Welcome to the Forum!
To help us analyze this issue we would need some additional details, such as whether you observe it across different models or only with a specific one, the full error message, payload details along with the code snippet, the prompts you are using, and the URL you are trying to access. Providing this information will enable us to reproduce the issue and work toward a solution.
Sure, thanks for the response.
Its tough to determine the error message as the response is simply empty inside of both the api and google studio. I have included an example site which I have attempted to read and the errors are either:
- The data is incorrect (likely outdated…could you confirm that the pages are not accessed real time but a previously indexed time that google did?)
- The response is simply empty. There just is no response. On our end, this is easier to replicate in studio. Sometimes rerunning 1-4 times can eventually produce a response.
I have tried many different prompts from something more complex to as simple as “read this url data and extract X” which still returns empty. Both flash and flash lite is having this issue.
One thing which would be great for you to confirm is that the data is based on last indexed time of the site, not the actual live url. For example, sites where data changes every minute would not be accessible or an older version would be. Ex. the site belows stock price.
stockanalysis com stocks/nvda/statistics/
Hello,
We would like to reproduce your issue in detail so that we can provide a more accurate and comprehensive response. For this, we would need your code along with the configuration and prompt. It would also be helpful if you could provide the complete response or complete error message.