*It’s interesting that the cost of prompts and platforms is becoming a major issue.<\/p>\n
From: The Prompt newsletter <\/p>\n
https:\/\/www.theprompt.io\/p\/write-great-llm-prompt<\/p>\n
I\u2019m seeing four things with folks who use LLMs in production right now:<\/p>\n
GPT-4 can do many things very well, but it can become quite expensive for production.<\/p>\n
Builders turn to GPT-3.5 Turbo for cost savings, but struggle with results;
\nThey then want to test models for other providers like Google and Anthropic, or an open-sourced model but they don’t get the same results.<\/p>\n
And lastly, they start thinking whether to fine-tune their own model.<\/p>\n
Let\u2019s talk about the strategies and how to go through this process.<\/p>\n
How to lower costs with GPT-4<\/p>\n
If you have a more complex LLM workflow, and you\u2019re using GPT-4 for every step of that workflow, the costs can start to rise.
\nTo reduce the cost, ask yourself this question:<\/p>\n
Which part of my workflow can I do with a secondary model?<\/p>\n
For example, GPT-3 will do great with summarization, but it won\u2019t do that well at capturing intent from a customer message. You\u2019ll definitely need GPT-4 for that.<\/p>\n
Depending on the use-case, you might end up using GPT-3.5 for more complex tasks. To do that, you need to create prompts that are similar to how the model was trained.<\/p>\n
In the next section we\u2019ll share some best practices that can improve your output.<\/p>\n
How to get better results with GPT-3.5<\/p>\n
Below are some tips to make GPT-3.5 work more like GPT-4:<\/p>\n
Use \u201cDo\u201d instead of \u201cDon\u2019t\u201d<\/p>\n
Separate Instructions from Context with ### or \u201c\u201c\u201c<\/p>\n
Be direct: Use \u201cYou must\u201d, or \u201cYour task is\u201d\u2026<\/p>\n
Assign a role<\/p>\n
Instead of just writing \u201cconversational style\u201d, add a sentence that follow that style, and ask the model to replicate that<\/p>\n
Add info about the end-user<\/p>\n
Provide the format structure of the output<\/p>\n
Give examples (use Chain of thought prompting for more complex reasoning tasks)<\/p>\n
Use emotion prompts like \u201cThis is very important for my career\u201d. (((Why does this work?)))<\/p>\n
If your prompt is complex, split it into more prompts and chain them together (it will be easier for GPT-3.5 to follow multiple but simpler prompts)<\/p>\n
Please refer to my original post for additional information and examples.<\/p>\n
How to prompt Claude?<\/p>\n
It\u2019s really interesting to me that we always default to prompting Claude the same way that we\u2019re prompting OpenAI\u2019s models. The reality is that this is a model that has been trained using a completely different set of techniques and methodologies, and requires a prompt design that adheres to those settings.<\/p>\n
Here\u2019s a quick rundown on how you should prompt Claude:<\/p>\n
Use XML tags like Use \u201cDo\u201d instead of \u201cDon\u2019t\u201d, and be direct<\/p>\n Start the Assistant output with the first token of the expected response<\/p>\n Assign a role, and pass it in the Assistant output as well<\/p>\n Ask Claude to reason through the problem before providing the answer<\/p>\n Provide Examples (few-shot, chain of thought prompting)<\/p>\n If you\u2019re dealing with longer documents, always ask your question at the end of the prompt<\/p>\n Break complex prompts into multiple prompts<\/p>\n If this seems a bit confusing, read my prompting guide for Claude, which includes detailed examples and instructions.<\/p>\n I\u2019m currently working on a prompting guide for Gemini and Mixtral, and will share it soon….<\/p>\n","protected":false},"excerpt":{"rendered":" *It’s interesting that the cost of prompts and platforms is becoming a major issue. From: The Prompt newsletter https:\/\/www.theprompt.io\/p\/write-great-llm-prompt I\u2019m seeing four things with folks who use LLMs in production right now: GPT-4 can do many things very well, but it can become quite expensive for production. Builders turn to GPT-3.5 Turbo for cost savings, […]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-4184","post","type-post","status-publish","format-standard","hentry","category-uncategorised"],"yoast_head":"\n