Do not keep bad results in context. You have to purge them to prevent them from effecting the next output. LLMs deceptively capable, but they don’t respond like

Do not keep bad results in context. You have to purge them to prevent them from effecting the next output. LLMs deceptively capable, but they don’t respond like a person. You can’t count on implicit context. You can’t count on parts of the implicit context having more weight than others.