agentgpt generates code that sometimes breaks due to word limits

[ouzhaoyuan]

[asim-shrestha]

Hey this is a symptom of limiting the token count. Not much we can do currently given its running off our own key, but something we'll look to fix in the future

[ouzhaoyuan]

Thank you. agentgpt It's very easy to use.

[LecJackS]

ChatGPT does not have the ability to reply more than 1 message in a row, but if the answer is incomplete, one can answer "continue" or "go on" so ChatGPT continues from the previous response.

In a general scenario (not sure if possible with AgentGPT) one workaround could be asking ChatGPT to output the word FINISHED at the end of each response, so if it's not there, we can reply with a "continue" to receive the continuation of its response.

Simple example:

Another way could be simply by replying "continue" at every long response (I think the limit is 4096 tokens) until a significantly shorter answer is given.