The simple prompt system is an unreleased tool for 1-way prompt testing and prompt template reuse. If there's solid interest I'll ship an alpha version for viewers of the channel. Like, sub, comment, @palidad1's comment etc to show interest.
I think a great follow up video would be one that discusses JSON in more depth and how to work with JSON that would be an important move. I understand JSON but it still has some mysteries that I would like to clear up thanks for a good video.
Content is Top! it is crazy these other channels have more subscribers than yours. Would you consider adding a follow up video? Top 6-10 techniques, like how important is roles and personas? compared to top 5?
It's hard to know without looking through the source. This is why I recommend staying 'close to the metal' and make sure you're the one writing the prompts. It's where all the value is.
I've been struggling to extract code from LLM output. I think EXAMPLES you mentioned would solve them. Thanks a lot! Can we use aider with non openai models? I couldn't make it work with Databricks models
I'm doing all of these without knowing it. The problem is I'm using small language model, sometimes it doesn't follow all the guidelines. I personally think you're 5 elements is great and really the most useful when doing prompt. But all 4 element is depending on the first one which is the model itself. So sometimes I can't make 1 single prompt like that. I must make it into 4 prompts so it follows all the guidelines ( Split the guide in 4 prompts so it is like step by step build up the result ). Anyway, thanks for making me realise there's pretty clear pattern there.
Yeah that will continue to be a problem. As mentioned, the model is element #1 and for a good reason. There's nothing more important than picking a high performing model. The time it takes to prompt a low performing model is NOT worth it. It's really REALLY not worth your time. Upgrade your model, spend a little, focus on results.
Absolutely BUT only after I've added these 5 elements will I touch roles/personas. Usually you just needed better static variables or examples. Roles/Personas are maybe element 7 or 8.
Soooo if whets most important to you is making the coolest "trick" then those are 5 important subjects, lets start out with you do fantastic work and are a far better programmer than i am. ... however if models so important , then how come someone that has done this everyday for years and someone sitting down for the first time are maxed out with the same model...i say the ability to train and skill up quicker is a far more important feature... you can always come in tomorrow and download the best model.... 1. automation, 2. being on the right side of the 'shaft you' line. 3. focus, 4. logic, and 5. time and i think you can wipe all 5 of those off the table honestly.
Context dependent. Some professions require reliability and consistency and hence more constraints. Others less so where the unpredictability is a plus. Example, imagine an agentic framework for surgery or taking 911 calls VS a children’s story writer….
There is no way that you have hand-written tens of thousands of prompts yourself. We haven't had AI and LLMs for long enough time for that to be true. I have written thousands of prompts, and I started in 2021.
honestly don't know why you've been stuck at ~14k subs for ages... content is gold. consistently.
What tool is that you are using for the prompt stuff
Yes! Really wanted to know that as well!
The simple prompt system is an unreleased tool for 1-way prompt testing and prompt template reuse. If there's solid interest I'll ship an alpha version for viewers of the channel. Like, sub, comment, @palidad1's comment etc to show interest.
Let me know how I can apply for this ? Thanks
Please share it
@@indydevdanany update on Ada ?
Model, purpose, variables, output example, JSON format.
Thank you very much 🙏
There are a lot of videos on RUclips about AI Agents and Prompts. Dan, your videos are spot on every time. Thank you, Sir!
Finally! Back to the good stuff. Much appreciated!
Good video. Shows new concepts to use in code, I especially liked the last 2 examples where it was all used together. Thank you IndyDevDan
I think a great follow up video would be one that discusses JSON in more depth and how to work with JSON that would be an important move. I understand JSON but it still has some mysteries that I would like to clear up thanks for a good video.
This video is gold!
Content is Top! it is crazy these other channels have more subscribers than yours.
Would you consider adding a follow up video?
Top 6-10 techniques, like how important is roles and personas? compared to top 5?
Pretty awesome indeed and great value / knowledge
What are you using for the box around your mouse cursor and text highlighting? It's really nice.
Beautiful. Appreciated, sir.
This content is gold. Thanks a lot!
Thanks for sharing, this is valuable alpha in the space!
Really enjoying your content, thank you. Like others, I’d like to know more about the tool you’re using to present in this video.
I am so sorry to ask, what platform is this? Or a Github repo?
Thank you this is one of the best videos
awesome, totally agree with these 5 elements. Do you think langchain or other libraries cover these elements well?
It's hard to know without looking through the source. This is why I recommend staying 'close to the metal' and make sure you're the one writing the prompts. It's where all the value is.
I've been struggling to extract code from LLM output. I think EXAMPLES you mentioned would solve them. Thanks a lot!
Can we use aider with non openai models? I couldn't make it work with Databricks models
Yes Aider supports opensource models: aider.chat/docs/llms.html
@@indydevdanPerfect. Thanks!
Do you find Personas add any benefit? There seems to be some research supporting them?
Can you talk about DSPy ?
I'm doing all of these without knowing it. The problem is I'm using small language model, sometimes it doesn't follow all the guidelines.
I personally think you're 5 elements is great and really the most useful when doing prompt. But all 4 element is depending on the first one which is the model itself.
So sometimes I can't make 1 single prompt like that. I must make it into 4 prompts so it follows all the guidelines ( Split the guide in 4 prompts so it is like step by step build up the result ).
Anyway, thanks for making me realise there's pretty clear pattern there.
Yeah that will continue to be a problem. As mentioned, the model is element #1 and for a good reason.
There's nothing more important than picking a high performing model.
The time it takes to prompt a low performing model is NOT worth it. It's really REALLY not worth your time. Upgrade your model, spend a little, focus on results.
@@indydevdan yeah.. I believe in you. 👍
In your experience, do you believe roles/personas add additional value?
Absolutely BUT only after I've added these 5 elements will I touch roles/personas. Usually you just needed better static variables or examples. Roles/Personas are maybe element 7 or 8.
Does all LLM model understand this !?
New to the channel, new to coding. What is this interface you are using?
Do you have any prompt repository to share? Is this omnicomplete prompt available anywhere or your own making?
Where is the prompt creation tools?
Soooo if whets most important to you is making the coolest "trick" then those are 5 important subjects, lets start out with you do fantastic work and are a far better programmer than i am. ... however if models so important , then how come someone that has done this everyday for years and someone sitting down for the first time are maxed out with the same model...i say the ability to train and skill up quicker is a far more important feature... you can always come in tomorrow and download the best model.... 1. automation, 2. being on the right side of the 'shaft you' line. 3. focus, 4. logic, and 5. time and i think you can wipe all 5 of those off the table honestly.
You digging GPT more
Than Claude?
Excellent Brother! I’m building #private AI federation servers. -R0N1N
Why is there so much desire to contain these LLMs? I see their value in the very opposite.
Context dependent.
Some professions require reliability and consistency and hence more constraints. Others less so where the unpredictability is a plus.
Example, imagine an agentic framework for surgery or taking 911 calls VS a children’s story writer….
There is no way that you have hand-written tens of thousands of prompts yourself. We haven't had AI and LLMs for long enough time for that to be true. I have written thousands of prompts, and I started in 2021.
10 cheeseburgers,100 cheeseburgers...doesn't matter.
Does the video have things folks can use.Thats the point.
Imagine applying #rop in a generalized manner. Allowing recursion.
en.wikipedia.org/wiki/Return-oriented_programming 🕯