JSON Parsing Error in generateObject Function with Unterminated String about ai

[lgrammel]

You have increased the temperature setting, leading to more random output, and are using a model that's weaker for tool calls (gpt-4o) in JSON mode. This is expected to happen, because the model might produce invalid JSON output in such a setting.

[pruett]

I tried this example with gpt-3.5-turbo and a temp of 0, and still receiving AI_JSONParseError. Even the example copy/pasted from https://sdk.vercel.ai/examples/next-app/basics/generating-object produces something similar.

Is this expected?

[lgrammel]

Please try without mode: json (then it uses tool calling w/ the OpenAI provider, which is often more robust). For better tool calling results, check out the prompt engineering with tools tips.

[pruett]

Thanks for the reply. No difference for me, even if I use generateText, for example. The error message is odd, too:

  cause: JSONParseError [AI_JSONParseError]: JSON parsing failed: Text: �webpack/lib/util/registerExternalSerializer�webpack-sources/RawSource�__webpack_require__.r(__webpack_exports__);

Clearly something is off given that it's parsing webpack 😅. All I'm trying is:

export async function GET(request: NextRequest) {
  // imports...

  const { text } = await generateText({
    model: openai("gpt-3.5-turbo"),
    prompt: "tell me a good joke.",
    temperature: 0,
  });
  
  console.log({ text });
  
  return new NextResponse();
}

I'm simply running this code within an API route. I must be doing something incorrectly. Just not sure what.

[pruett]

Problem solved ✅

Solution: Changing the API route from a GET to a POST

I'm guessing there were weird caching issues from marking the route as a GET that goes away when switching to a POST. Hope that this helps anyone that might stumble across this issue.