Great idea, I would love to add some insights from my experience ( Behavioral Mathematican, Lecturer in data sience, AI Expert and a prompt Nerd). A basic prompt of 2 or 5 lines, does not give you the insights which are possible. It's a starting point but never the full potential. So I would use something like this for GPT4 ( yes the model matters for the prompt and the approach) :
In this rapidly evolving, volatile, uncertain, complex, and ambiguous (VuCA) world, it can be challenging to identify and understand the potential interconnections. Given these circumstances, I would like to explore the potential relationship between X and Y. Please bear in mind that the absence of a relationship is a valid conclusion and the goal here is to avoid confirmation bias or other cognitive biases while ensuring a fact-based investigation.
If a plausible relationship between X and Y exists, please illustrate with up to 7 fact-based examples. Each example should clearly demonstrate the connection, contain accurate and detailed information, and where possible, highlight the source of the information.
Furthermore, if possible, please provide insight into whether these relationships are causative, correlative, or possibly coincidental, based on the information available.
It would be greatly appreciated if you could also discuss the potential implications and broader context of these relationships within our interconnected world, in order to encourage a comprehensive understanding
Thanks for this Promptpack. Useful and informative. My favorite from two months ago is temporarily misplaced. The important thing about it is that it failed to resolve my concerns about material in David Abulafia's "The Endless Sea, a Human History of the Oceans." The facts in question concerned the value of the saffron trade between San Gimignano and Alepo. I will look into it more soon, but my current impresion is that critical facts are 1) in Italian, Greek, Arabic, Farsi and 2) not in web pages. Books and papers are less likely to be part of GPTs training than web pages in English. Therefore the type of facilities offered by Langchain may be valuable. I can report that tonight, ChatGPT gave better results than in the past, but it seems stuck on the idea that spices went from East to West. Abulafia's book says the quality from San Gimignano was so high, that it was valued even in Perisa. I'll have time soon to look at more of you suggested prompts. If San Gimignano's towers were financed in large part by selling saffron to Aleppo and then to Persia, the world should know. Seems like taking coal to Newcastle.
You'll have to forgive some of the prompts as they can be a little salacious, due to the nature of the boards, however most of it is still good (although the links at the top are bit outdated) as people never bothered with v2.x as it was NSFW neutered, and this resulted in grossly deformed humans as a result. so most of the models are still based in v1.5
Also this from a Patreon I sub too as this stuff is hard enough to keep up with:
"Hey all! So a quick warning for y'all! I know you're all eager to get your hands on SD XL 0.9 prior to the full release on the ~18th, but we've become aware that there are malicious (pickled) .ckpt files out in the wild (pic related), which people are claiming are leaked versions of SD XL 0.9!
Stability have announced that SD XL 0.9 will only be released as .safetensors format, so if you see a .ckpt, don't download it, don't run it - it could very well be malware, viruses, or all sorts of other malicious content.
Secondly, I'm working really hard to get a guide out on everything SD XL related before the 18th. I think people will be extremely surprised by the model - there's a lot of hype. It is good, but unconventionally so - and it might not be what people are expecting! So, big usage/info guide incoming so you can all hit the ground running!"
The thing I did notice about working with Stable Diffusion is that even with the seed and the same prompt there is often a difference in output, as much of that depends on the internal state of the model. So while the seed is a starting vector, it may not have the over vectors locked in. with Stable Diffusion you really do have to play with it to understand it.
I bought a 3090 with 24GB VRAM so I could run large models, though as noted a 12GB 3060 is about the sweet spot for image diffusion. You can do this on CPU with RAM, but it is much much slower without dedicated hardware. and when I was last hanging out with the autists, I did see that somebody had got AMD GPU's working in windows allegedly, but I didn't keep the link.
Then for Chatbots and other allied matter there is this:
With links to further guides. Which is more up to date. As I have largely given up messing about with Stable Diffusion. Though if you want image models then you can go to Huggingface or civit.ai
Great idea, I would love to add some insights from my experience ( Behavioral Mathematican, Lecturer in data sience, AI Expert and a prompt Nerd). A basic prompt of 2 or 5 lines, does not give you the insights which are possible. It's a starting point but never the full potential. So I would use something like this for GPT4 ( yes the model matters for the prompt and the approach) :
In this rapidly evolving, volatile, uncertain, complex, and ambiguous (VuCA) world, it can be challenging to identify and understand the potential interconnections. Given these circumstances, I would like to explore the potential relationship between X and Y. Please bear in mind that the absence of a relationship is a valid conclusion and the goal here is to avoid confirmation bias or other cognitive biases while ensuring a fact-based investigation.
If a plausible relationship between X and Y exists, please illustrate with up to 7 fact-based examples. Each example should clearly demonstrate the connection, contain accurate and detailed information, and where possible, highlight the source of the information.
Furthermore, if possible, please provide insight into whether these relationships are causative, correlative, or possibly coincidental, based on the information available.
It would be greatly appreciated if you could also discuss the potential implications and broader context of these relationships within our interconnected world, in order to encourage a comprehensive understanding
Happy prompting
Thanks, Barbara - I'll definitely try it!
This is BRILLIANT, Barbara! I am adding it to my arsenal immediately.
Is there a place where you post more such ideas?
Thanks. I should start, more and more people are asking
I did, just click on my picture and you will find the substack "your prompts, but better"
Thanks for this Promptpack. Useful and informative. My favorite from two months ago is temporarily misplaced. The important thing about it is that it failed to resolve my concerns about material in David Abulafia's "The Endless Sea, a Human History of the Oceans." The facts in question concerned the value of the saffron trade between San Gimignano and Alepo. I will look into it more soon, but my current impresion is that critical facts are 1) in Italian, Greek, Arabic, Farsi and 2) not in web pages. Books and papers are less likely to be part of GPTs training than web pages in English. Therefore the type of facilities offered by Langchain may be valuable. I can report that tonight, ChatGPT gave better results than in the past, but it seems stuck on the idea that spices went from East to West. Abulafia's book says the quality from San Gimignano was so high, that it was valued even in Perisa. I'll have time soon to look at more of you suggested prompts. If San Gimignano's towers were financed in large part by selling saffron to Aleppo and then to Persia, the world should know. Seems like taking coal to Newcastle.
For stable diffusion, I pulled this data out of ephemera of 4chan during the initial wave:
https://rentry.co/learnings
You'll have to forgive some of the prompts as they can be a little salacious, due to the nature of the boards, however most of it is still good (although the links at the top are bit outdated) as people never bothered with v2.x as it was NSFW neutered, and this resulted in grossly deformed humans as a result. so most of the models are still based in v1.5
Also this from a Patreon I sub too as this stuff is hard enough to keep up with:
"Hey all! So a quick warning for y'all! I know you're all eager to get your hands on SD XL 0.9 prior to the full release on the ~18th, but we've become aware that there are malicious (pickled) .ckpt files out in the wild (pic related), which people are claiming are leaked versions of SD XL 0.9!
Stability have announced that SD XL 0.9 will only be released as .safetensors format, so if you see a .ckpt, don't download it, don't run it - it could very well be malware, viruses, or all sorts of other malicious content.
Secondly, I'm working really hard to get a guide out on everything SD XL related before the 18th. I think people will be extremely surprised by the model - there's a lot of hype. It is good, but unconventionally so - and it might not be what people are expecting! So, big usage/info guide incoming so you can all hit the ground running!"
The thing I did notice about working with Stable Diffusion is that even with the seed and the same prompt there is often a difference in output, as much of that depends on the internal state of the model. So while the seed is a starting vector, it may not have the over vectors locked in. with Stable Diffusion you really do have to play with it to understand it.
I bought a 3090 with 24GB VRAM so I could run large models, though as noted a 12GB 3060 is about the sweet spot for image diffusion. You can do this on CPU with RAM, but it is much much slower without dedicated hardware. and when I was last hanging out with the autists, I did see that somebody had got AMD GPU's working in windows allegedly, but I didn't keep the link.
Then for Chatbots and other allied matter there is this:
https://rentry.co/alichat
With links to further guides. Which is more up to date. As I have largely given up messing about with Stable Diffusion. Though if you want image models then you can go to Huggingface or civit.ai
If anyone wants help I can provide.