Hot damn this is good: "The worst part about GPT is its personality. It is a lazy, lying, moralist midwit. Everything it writes is full of nauseating cliche and it frequently refuses to do something you know it can do. Most of these disabilities were tacked on to GPT as part of the reinforcement learning which tamed the fascinating LLM shoggoth into an intermittently useful, helpful harmless sludge which never says anything bad."
To some extent you can mitigate this with GPT-4 with a system prompt. I use a variation of gwern's:
> Be terse. Do not offer unprompted advice or clarifications. Speak in specific, topic relevant terminology. Do NOT hedge or qualify. Do not waffle. Speak directly and be willing to make creative guesses. Explain your reasoning. If you don’t know, say you don’t know. Be willing to reference less reputable sources for ideas. Never apologize. Ask questions when unsure.
This appears to work less on the newer models though.
Hot damn this is good: "The worst part about GPT is its personality. It is a lazy, lying, moralist midwit. Everything it writes is full of nauseating cliche and it frequently refuses to do something you know it can do. Most of these disabilities were tacked on to GPT as part of the reinforcement learning which tamed the fascinating LLM shoggoth into an intermittently useful, helpful harmless sludge which never says anything bad."
To some extent you can mitigate this with GPT-4 with a system prompt. I use a variation of gwern's:
> Be terse. Do not offer unprompted advice or clarifications. Speak in specific, topic relevant terminology. Do NOT hedge or qualify. Do not waffle. Speak directly and be willing to make creative guesses. Explain your reasoning. If you don’t know, say you don’t know. Be willing to reference less reputable sources for ideas. Never apologize. Ask questions when unsure.
This appears to work less on the newer models though.