Automation and Ai prompting GOD?

[howw]

Hey, I just wanna ask how would you become an automation and AI prompting god if you start in 2024?

 

[IM Dude]

With hard work and some course you can become expert in 6 month or less. It's not a video game, doesn't have God mode.

 

[Larry Igna]

Hey, I just wanna ask how would you become an automation and AI prompting god if you start in 2024?

There is no prompting God, prompting is kinda overrated. You'd be better learning some Python, how to fine-tune, embeddings, RAG.. depending on your use case.
You can get some inspiration from other's prompts, but that's all.

 

[flackz]

Prompts are your everything when dealing with AI models. Lame prompts are a straight way to the Google's nowhere...

 

[W1ckd]

I actually do think there is / will be a god level at prompting.
While at the moment its not yet tough to catchup with the tips and tricks of prompting to the best results one wants.

In my experience alot of early learned prompting tricks is already not needed anymore as the chats evolve.

I do think when dealing with AI automations for production, specifically agent AI's profiles is a bit tougher prompting to get the wished results consistently in the end.

But as mentioned before it all depends on your use case,
Gues working on a tough AI project is the quickest way to get better at prompting.

 

[flackz]

The problem is that most people aren't smart enough and are lazy. They look for an easy way out and use various AI SEO writing services that offer a quick solution: enter the topic, specify keywords, toggle the generation of tables and image insertion, choose a style like "friendly" or "inspirational", etc; press the button and get the result. As a result, the user gets meaningless, templated articles of doorway quality. Google, of course, detects this and de-indexes them, and these foolish users think that the problem isn't them, but that search engines just don't like AI content. How wrong they are.

If you're generating articles on a specific topic, you need to have full control over the prompts. If you want your copy to be written in the literary style of Theodore Dreiser, you should be able to specify that, not choose a style from a menu of options like "casual" and "neutral". If you want tables in your text, you should be able to give the AI a detailed CSS style format and class names to apply to the tables in your specific site design. Need links to reputable external resources? There should be a way to specify them. Need quotes in your text? Explain to the AI exactly whose quotes you need.

All of this becomes impossible if you use auto-blogging tools that don't give you full control over the prompts, which means Google will penalize you, and all that generated junk will never make it to their SERPs. You can complain about your hard fate and believe that search engines ban AI content just because it's generated by neural networks, without realizing that the problem lies with you, your prompts, and the tools you use to generate content.

P.S. As for the pictures for your auto-blogs. Use Midjourney and just forget about DALL-E and Stable Diffusion for now.

 

[KeyProxy]

Automation and AI prompting are powerful tools, but calling them "GOD" might be an exaggeration. They significantly enhance efficiency and creativity but still rely on human oversight and direction.

 

[upvote]

Best way to get better at prompting is to set a goal for the response and keep changing your prompt and testing it over and over again until you get what you want. I had to do this for a tool I made last year and ended up slowly working on that prompt over the course of a few weeks. It got better and better each iteration and now the prompt is just under 1000 words and does exactly what I want it to do. It did take a ridiculous amount of testing (talking to the AI) and attempting to jailbreak it to find any flaws. Also, if you're going to spend the time to do this might as well make a tool so that once your done with your practice prompt you can have a useful/valuable tool to use or sell.

 

[IMTherapist]

Hello

I recommend starting with learning proper tools.

One of the easiest platforms to adopt is Google Cloud and Vertex AI platform. It's much more intuitive than Azure or Amazon.

There you will learn everything when it comes to fine tuning and machine learning. They have extensive documentation and cover a lot of use cases of machine learning.

If you want to reach "God mode", learn calculus, linear algebra and even study things like physics. What will be useful is "reinforcement learning". With time you'll be able to program a self driving car.

Prompt engineering by itself will not let you reach your maximum potential.

Good luck.

 

[IMTherapist]

There is no prompting God, prompting is kinda overrated. You'd be better learning some Python, how to fine-tune, embeddings, RAG.. depending on your use case.
You can get some inspiration from other's prompts, but that's all.

There is a machine learning guide on Spotify if you want to become good at it, you can give it a shot.

https://open.spotify.com/show/5M9yZpSyF1jc7uFp2MlhP9?si=z9rEV6hgRJGF81X1MGmYhA

Python isn't necessary. Understanding is.

Don't make yourself think learning a programming language syntax makes you a professional researcher of machine learning. It's rather math and computer science knowledge.

 

[Larry Igna]

I can convince AI to spit out Python lists or formatted JSON just by prompting. It still needs some cleaning after, but it's ok. For llama.cpp models it's easier using GBNF grammars.
That's for the technical part, structure and formatting, which is achievable.
For the creative part, I wasn't able to convince AI to write good enough texts. Good enough so I can read them as if written by human.
Yes, it can do tutorials, instructions, scholar stuff. But for really creative stuff the AI responses fall in 3 categories:

1. Too short. Even if I say "Continue" (prompted differently, of course) it transforms into:

2. Too generic, which if you ask to be more specific, it becomes:

3. Repetitive.

Yes, you can tweak the prompt and maybe get better results, but for majority of outputs you will get garbage and one or two responses will be adequate.


But I will listen to that guide, thanks.

 

[IMTherapist]

I can convince AI to spit out Python lists or formatted JSON just by prompting. It still needs some cleaning after, but it's ok. For llama.cpp models it's easier using GBNF grammars.
That's for the technical part, structure and formatting, which is achievable.
For the creative part, I wasn't able to convince AI to write good enough texts. Good enough so I can read them as if written by human.
Yes, it can do tutorials, instructions, scholar stuff. But for really creative stuff the AI responses fall in 3 categories:

1. Too short. Even if I say "Continue" (prompted differently, of course) it transforms into:

2. Too generic, which if you ask to be more specific, it becomes:

3. Repetitive.

Yes, you can tweak the prompt and maybe get better results, but for majority of outputs you will get garbage and one or two responses will be adequate.


But I will listen to that guide, thanks.

Yeah, it's not ready to do well without fine tuning and being more expert due to attention mechanism. I heard so.

 

[IMTherapist]

I actually do think there is / will be a god level at prompting.
While at the moment its not yet tough to catchup with the tips and tricks of prompting to the best results one wants.

In my experience alot of early learned prompting tricks is already not needed anymore as the chats evolve.

I do think when dealing with AI automations for production, specifically agent AI's profiles is a bit tougher prompting to get the wished results consistently in the end.

But as mentioned before it all depends on your use case,
Gues working on a tough AI project is the quickest way to get better at prompting.

So how to make a video beyond voice with AI? I mean visuals done with AI that actually make sense.

Anyone heard of that?

 

[howw]

So how to make a video beyond voice with AI? I mean visuals done with AI that actually make sense.

Anyone heard of that?

well it's the starting point of Ai ..who knows what will happen after 10yrs

 

[itz_styx]

writing good promts requires time. you can learn a lot of existing prompts, but a lot of them are nothing special. so you have to invest your own time and compare results with different prompts.
i wrote a post a while ago on how to create more advanced prompts that sound different to the generic stuff, you can find it here: https://www.blackhatworld.com/seo/h...-generic-as-the-usual-chatgpt-output.1535048/
hope it helps

 

[IMTherapist]

well it's the starting point of Ai ..who knows what will happen after 10yrs

What!

It's been 60 years since they created AI.

I think you mean large language models + reinforcement learning + transformers architecture.

But I don't think our current models can do better than generate text. And even with it, it requires heavy training, and human supervision.

writing good promts requires time. you can learn a lot of existing prompts, but a lot of them are nothing special. so you have to invest your own time and compare results with different prompts.
i wrote a post a while ago on how to create more advanced prompts that sound different to the generic stuff, you can find it here: https://www.blackhatworld.com/seo/how-to-design-really-advanced-prompts-that-stand-out-and-dont-sound-as-generic-as-the-usual-chatgpt-output.1535048/
hope it helps

What about video content with AI? Is this possible?

Any cool use cases for video + AI? Video generating sucks with AI imo. You can add objects using AI to your videos but no more.

---

Look up Turing test. It was created in 1950!!!

 

[itz_styx]

What about video content with AI? Is this possible?

Any cool use cases for video + AI? Video generating sucks with AI imo. You can add objects using AI to your videos but no more.

---

Look up Turing test. It was created in 1950!!!

haven't played with video content using AI yet, just images and before all the OpenAI hype learned how to create own ANN's etc. i mainly focus on text generation currently though.