• Unreal AI
  • Posts
  • 6 prompting strategies for better results from Open AI

6 prompting strategies for better results from Open AI

Author

Alie Jules
December 18, 2023

Things seemed to be winding down for the holidays, well, except for the hype about GPT 4.5. And if we are lucky, we may get Midjourney v6 just before the holidays.

Let’s go!

The Latest

CHATGPT 4.?
ChatGPT 4.5 hype has been in overdrive on X for the last couple of days. Especially since some users managed to get ChatGPT to output that it’s running on GPT 4.5 turbo?!?!

Only hype (I think).

MIDJOURNEY
Midjourney is getting closer to the v6 release! Current users are now doing the first rating party, rating image pairs, which is one of the last steps before v6 release! Midjourney did a poll on Discord asking people whether they preferred a release of v6 ASAP or on Jan 1st. The winner? ASAP!

DIGI AI
We have reached a new level of AI, sort of, in the AI romance department. A post on X from the creator of Digi, the future of romantic companionship, has garnered over 21 million views. Would you want to have an AI romantic partner?
For me, it’s a definite no.

OPEN AI
Open AI released its prompt engineering guide with six strategies for better results.

You may have seen/used some of these, but it’s great to get this directly from Open AI.

  1. Write clear instructions - models can’t read your mind!

  2. Provide reference test - helps reduce made-up/fake answers

  3. Split complex tasks into simpler subtasks - complex tasks tend to have higher error rates

  4. Give models time to “think” - ask the model to “think it through” before rushing to answer

  5. Use external tools - if a task can be done better by a tool vs. an LM offload it to get the best of both

  6. Test changes systematically - evaluate model outputs with reference to gold-standard answers

You can combine these, too!

The prompt engineering guide includes detailed tactics and examples as well.

TINY MLs
Scientists from MIT and University of California campuses say LLMs can now create smaller AI tools without humans.

"Right now, we're using bigger models to build the smaller models, like a bigger brother helping [its smaller] brother to improve.”
Yan Sun, CEO of Aizip

The Tiny MLs focus on specific, individual/small tasks - and at a much lower cost.

BYTEDANCE
ByteDance has been using OpenAI's technology (behind the scenes) for about a year to build its own LLM competing model, codenamed "project seed." This is a direct violation of OpenAI's and Microsoft's Terms of Service. According to Alex Heath, who broke the story: "The misuse is so rampant that Project Seed employees regularly hit their max allowance for API access."

OpenAI has suspended ByteDance's account while they investigate.

Meanwhile…

Make money with AI? Sure! An Indiepreneur crossed $10,000 for the year ($1000+/mo) with an AI product that, according to him, “doesn’t solve a problem”. How it works: send in a picture of you and your partner (future parents) and get “your future baby” (boy and a girl) images generated by AI. Simple concept. OurbabyAI.

That is it for today’s AI news roundup. I will leave you with this prompt.

closeup of freshly baked gingerbread cookies on a plate, festive atmosphere --ar 3:2

Until Wednesday,

Alie