Prompt Engineering: Use AI to create your magnum opus
Learn new prompt-engineering techniques for the best AI outputs—from art to writing!
Welcome (back) to AI Minds, a newsletter about the brainy and sometimes zany world of AI, brought to you by the Deepgram editorial team.
In this edition:
🌲 Advanced Prompt Engineering Techniques (e.g. Tree-of-Thought)
🎨 Prompt engineering techniques for AI art (e.g. weights)
🔨 Prompt engineering’s impact on jobs (Paper included)
📚️ RAG + Prompt Engineering (Twitter thread)
😮 Malicious prompts and LLM Jailbreaking (e.g. DAN)
📈 Bonus content! Distillation via prompting, FrugalGPT, and more!
We at Deepgram make language models, too! Here’s a one-minute video about our latest (and greatest) one.
🚀 Advanced Prompt Engineering Techniques
Chain-of-Thought Prompting is an approach that encourages LLMs to break down a complex “thought” into intermediate steps by providing a few demonstrations to the LLM (aka few-shot learning). Learn by example above.
Tree of Thought Prompting is supposed to mimic something like human brainstorming—creating and considering diverse ideas. There are four steps to “Crafting and LLM Tree,” including a BFS- or DFS-like approach. Learn more above:
🖼️ Crafting Masterpieces: Prompt Engineering Art
The video above will take you through various prompt-engineering techniques, from simple instruction prompting to more advanced self-consistency decoding.
Fun fact: The thumbnail art was made with the prompt engineering techniques Jose discusses. And the video itself walks through this thumbnail creation process.
If you’d rather read than watch, check out the article below.
Whether you’re using AI to write an article or to create eye-catching images, your output can only be as good as your prompt.
🛠️ Will prompting become my job?
Check out the quote below from this article! Nithanth gives an overview of the impact of prompt engineering at large.
🐦 Tweet alert: RAG + Prompt Engineering
Last week we talked about RAG. It turns out you can use prompt engineering to optimize RAG. Check out this tweet!
When should you do prompt engineering to optimize RAG? 📝
Answer: it should probably be one of the first “table-stakes” things you do to tackle failures like LLM hallucination.
We’ve added a full Prompt module guides along with a wealth of resources on how to tackle prompts in… twitter.com/i/web/status/1…
— LlamaIndex 🦙 (@llama_index)
Oct 31, 2023
🏎️ Jailbreaking: When Prompt Engineering Turns Malicious
Malicious Prompts: Certain users have invented prompts like DAN (“Do Anything Now”), to bypass safety protocols and lead ChatGPT to give some mischievous (though effective) advice. To see these rapscallion prompts, check out the blog below!
🤖 Other bits and bytes
Before prompt engineering, you may end up with the image on the left. After learning some techniques, you can end up with the image on the right 🤩
🧪Distillation through chain-of-thought prompting - It turns out you can distill an LLM through Chain-of-Thought prompting. This paper breaks it all down.
💰FrugalGPT - Prompting LLMs isn’t free. This paper analyzes which prompts to send to which LLMs such that users can optimize their own personal economic efficiency when using AI.
🤯DAN 8.0 reddit thread - The “evil genius” prompt is thoroughly discussed in this Reddit thread. It’s 517 words long. Check it out!
Side note: In our video on Llama-2, we showcase just how easy it is to get a new LLM (Llama-2-chat) to hallucinate and output factually incorrect responses.
Fun fact: Every piece of art in this edition of AI Minds is AI-generated!