You should look at the consumer app system prompt: twitter.com/amandaaskell/status/1765207… I've had way more success taking it and forming it to my tasks than stuffing in a bunch of tokens about how to do things. In one case I was able to go from ~6k in input tokens to ~3k because I no longer had to provide a mountain of examples and instructions for corner cases

source: twitter.com/amandaaskell/status/1765207…

Random, off-topic observation: it seems like I see lots of open-source LLM prompts have prompts that aren't proofread. For example [0], "ounch", "he spend", general lack of punctuation, and so on. But the big model providers don't do this [1], so I'm curious how much impact proper grammar in prompts makes (and if it's a positive or negative impact). [0] github.com/allenai/lumos/blob/main/data…... [1] twitter.com/AmandaAskell/status/1765207…

Claude's system prompt is given here -- they're not trying to hide it: twitter.com/AmandaAskell/status/1765207…... It doesn't actually include that text, but it may have been trained in. (Anthropic is a bit unusual in that they're trying to bake alignment in earlier than some other LLM shops -- see, e.g., anthropic.com/news/claudes-constitution

Keep in mind that this is not the only way the experts program this technology. There's plenty of fine-tuning and RLHF involved too, that's mostly how "model alignment" works for example. The system prompt exists merely as an extra precaution to reinforce the behaviors learned in RLHF, to explain some subtleties that would be otherwise hard to learn, and to fix little mistakes that remain after fine-tuning. You can verify that this is true by using the model through the API, where you can set a custom system prompt. Even if your prompt is very short, most behaviors still remain pretty similar. There's an interesting X thread from the researchers at Anthropic on why their prompt is the way it is at [1][2]. [1] twitter.com/AmandaAskell/status/1765207…?... [2] and for those without an X account, nitter.poast.org/AmandaAskell/status/17…...