I am encountering two errors that did not previously exist in a completion prompt I’m running:
The response back says that Open AI cannot browse the Internet (“I’m unable to access external websites directly.”). When I take the prompt and put it into ChatGPT it works fine. It appears that something must have changed with how the API is being called.
I get a max token error when I have 0 set. 0 should be the indicator to use the max for the model.
I have tried my step with each GPT model with the same results, except with #2 which seems to only affect the new models.
Has something changed with the API that Make.com needs to fix?
Sadness, @asmang. How long ago did the problem arise?
I confirm that the OpenAI module doesn’t check a website. This is expected behavior for me, because I’m used to needing to scrape a site before using it as OpenAI source materials through the API/modules like this.
It should and it has for me. This is recent, maybe the last week or so? I set the project aside for a little bit.
I’m getting the same response as you but wasn’t previously. I will have to also try it in the OpenAI sandbox, though I’ve seen it work there before and feel like there’s a module update needed by Make.com since there have been a few API updates.
No, but I believe you may have within my note about ChatGPT. What I was saying was:
Without making any changes to my previously working module, it now provides an error response (not an API message but a result with text saying it can’t browse).
If I take the generated prompt and then put it in ChatGPT it works.
I have credit in my OpenAI account.
Prompt snippet (this had been working fine) is below. Note that I’ve tried gpt-4 as the model as well and that didn’t work but worked fine in the OpenAI playground.
P.S.: Did you know, the concepts of about 70% of questions asked on this forum are already covered in the Make Academy. Investing some effort into it will save you lots of time and frustration using Make later!
Hello @asmang, @samliew is right about the GPT-4o model and other models via the OpenAI developer platform do not have access to the internet.
I also knew that at this moment using API(not interface) there is no one( popular AI Model) able to extract live data from the Internet.
I’ve also read multiple articles about this in different communities where I’ve found that Perplexity can able to do that but It’s not 100% accurate because it’s ‘pre-indexed’ from a WebCrawl (like what Google does) and not live lookups however. Official community post from OpenAI
It seemed to be previously working. I can try Perplexity, but it’s very odd that there’s such a huge change all of a sudden if it was never working. Any thoughts? I appreciate you and @samliew replying.
These AI Models are pre-trained based on previously available data.
I’m not 100% sure but maybe that request you asked for which is already there on their previously trained data.
Another important factor is I’ve asked directly GTP for their model cut-off dates.