GPT Prompt Strategy: Brainstorm, Search, Hypothesize, and Refine - THIS is the FUTURE!!

Patreon (and Discord)
/ daveshap
Substack (Free)
daveshap.substack.com/
GitHub (Open Source)
github.com/daveshap
AI Channel
/ @daveshap
Systems Thinking Channel
/ @systems.thinking
Mythic Archetypes Channel
/ @mythicarchetypes
Pragmatic Progressive Channel
/ @pragmaticprogressive
Sacred Masculinity Channel
/ @sacred.masculinity

[@goatnamese]

The gods have favored me. I was just starting to scrape my bother's channel manually today. 250 videos of 1 hr each. My tears have cleared up. Thanks so much!!!

[@DaveShap]

Share what you find on the GitHub! I can use some help

[@llN3M3515ll]

"Just don't do anything.. creepy", challenge accepted! 😂

[@tomadd8165]

glad to see you're back with AI-hacking videos! These are the best... applying high quality theory and insights from neuroscience, librarian theory, creative writing, etc to practical uses of AI... keep those coming... that's how i discovered your channel, that's what makes your channel so different, that's why i'm eagerly watching new videos...

[@luiswebdev8292]

those last videos knock it off the park!

[@mc9723]

Love you and Matt's work! Thanks to both for keeping us informed!

[@DaveShap]

Yeah, no shade for my fellow creators! Corporations on the other hand...

[@haileycollet4147]

Love that it's all separated out. Makes it easy to use different models for different steps (e.g. generating search queries, especially the first time, with 3.5)

[@Thebentist]

Can’t wait to see how this turns out! Excited to try it myself!

[@DaveShap]

Oh hey, good to see you :) yeah stay tuned, will be working on this to get it fully implemented at least as an MVP/POC

[@J3R3MI6]

Dave’s content and code is well worth supporting 🙏🏽

[@goforit5]

Informative, insightful content as always. I learn so much from you Dave. Thanks!

[@chitdigital7057]

Great video thanks. Im new to tech and AI and all this gets overwhelming but you make it pretty clear.

[@omshantiom-]

ill support you. your content is great. thanks for all you do!

[@1-chaz-1]

Love this, nice work man!

[@xox14]

Thank you for these amazing videos!

[@godned74]

This stuff simply works. Great work man.

[@jakubdabrowski7774]

first :) Love your videos Dave, thanks for sharing your great ideas.

[@DallasMcMillan]

Hey Dave! I've always been a big fan of your videos, and I recently checked out your new video testing "more professional" stock animations. I have to say, like many others who commented on that post, I find your regular format to be more enjoyable and educational for several reasons . The functional and intelligent imagery that complements the content is a big plus (unlike the stock footage, which felt like filler - that's the opposite of AI if you think about it). We also really appreciate the red shirt and the other sci-fi and nerd references you incorporate. But most importantly, what sets your content apart is its authenticity, genuineness, and unapologetic human touch. You truly care about the mission and its implications, and it resonates with your audience, making us feel connected and like we "get it" too. You're ticking all the right boxes for creating truly exceptional content and fostering genuine connections through digital media. Keep up the fantastic work!

[@mbottambotta]

the thing that tends to happen to me when I watch David Shapiro's post is that I want to press the Like button more than once.

[@boldandthebeautifulgimbal2881]

11:10 I came up with ‘Focalyximg’ - The act of intense vision focus, with an intent to analyze sensory input, and interpret.

[@jonathanlindsey8864]

O yhea Happy "birthday" to your KZbin channel! 🎉

[@jorgetonos]

Huge fan here... I def want to try this.

[@caseyhoward8261]

You are amazing, my guy.

[@RafiDude]

Thanks!

[@officetunes7295]

Great video!!

[@Will-kt5jk]

In my model (still not implemented in years of thinking) there’s effectively a more formal “Test” step between “hypothesise” & “refine”. This was originally considering human processes, so I was also considering branching, which sort of maps to tree of thoughts in LLM world.

[@DaveShap]

Yeah the test is the satisficing. Did you have something else on mind?

[@PizzaLord]

One of the more interesting questions is that, despite all this advance, why is there no such thing as an AI that can teach a syllabus? After all a syllabus is a bunch of written info and concepts that are linked. It also has an order and dependencies like if you are 20 percent the way teaching something and a student asks something that you were planning to teach at the 80 percent mark that depends on concepts you introduce at the 50 percent mark that you have not arrived at yet then that's a dependency. Why does this not exist and if it does, what is the best AI at doing this right now? And if it does not exist, how would you build it using the current tech we have? I am serious about this and I am a tech architect and coder so if anyone wants to team up, let me know.

[@JakexGaming]

That's a really cool idea

[@actualityai]

@@JakexGaming It feels like LLMs are not necessarily outright designed for just sitting there as a corpus of knowledge and then going "ok, ask me a question about anything you like" but that it what we are using them for right now in terms of the frameworks and tools being developed around them.

[@8eck]

In any case, i'm intrigued by your ideas. I'm currently also thinking and working over a R&D project, to perform some kind of semantic compression and summarization of large amount of information in an iterative manner and without dramatic/critical information loss.

[@DasRaetsel]

You made a good choice in marrying a librarian. Great insights!

[@ilanapregen6010]

i like that you used “tldr” correctly. i have heard it used so many times to mean “the short of it is…” Someone will say, basically, “it’s a long story; but the tldr is (insert shortened version of whatever happened).” What people mean to say in that circumstance is, “the story is tldr, but the short of it is (insert shortened version of what happened).”

[@rasmusfoy]

Immense gratitude for your content. I just discovered your channel and am loving your ability to break things down so even an intermediate like myself can follow. I am spending my free time doing the deep dive in to what is possible to build with generative AI. I decided to pivot my career in business and real estate to follow my passion for emerging and exponential tech. Any advice on where to start beyond learning python 😂? I studied CivEng.

[@StephenRayner]

❤ amazing video

[@candlespotlight]

I'd love to see more about BSHR, latent space activation, and search query stuff in general (like it would be incredible to see something that is actually using AI to its FULLEST for finding information, like you were talking about being possible)!! Those were among my favorites from you!! I'd love for you to keep going with those things and for others to join in as well, but I know it's impossible to focus on many things at once, and you have a lot on your plate. I am not qualified enough to contribute, but I am an interested amateur, and so I can provide feedback on anything if that would be helpful.

[@xox14]

The closing background song 💯

[@lukeaussie08]

Thanks

[@DaveShap]

Thank you! 😁🖖

[@xox14]

"its probably gonna be three or four part series" the famous last words 🤣🤣

[@yonatan09]

Wow, that's a great idea will save a lot of time in the long end. Didn't know you had written books. I already downloaded the last one.😊 I am yonatan from ethiopia.

[@DaveShap]

Hello in Africa!

[@yonatan09]

Hey bro, I recently discovered your channel and all the videos are well made. I always seem to finish them. I think you spend a good time on your slides

[@yonatan09]

​​@@DaveShapwhere do you stand on Emad Mostaque's idea of "one single AGI overtaking everything vs Swarms of AI augmenting human intelligence " ??? Would love an answer, maybe even a video

[@TarninTheGreat]

Great video so far! Paused at 15:00ish, Not only that all, but also: How do you define power? The answers it gave you yesterday were about reformers and philosophers trying to steer Rome away from it's horribly violent genocidal fasc-ist tendancies. Cicero, Aurelius. It didn't say antony. It didn't say augustus or crassus or nero. It gave you an answer that included what it thought power was, or greatness, or peakness, or whatever. Which doesn't seem in line with your unstated assumptions in the question. As for me, the senators at the height of rome are probably the ireni or sabine chiefs who told the sea peoples after the folly of troy to 'please don't settle here, you're far too violent, go back to Dido, she actually loved you... . oh she killed herself because you were an asshole.... eesh, *never be a friend to rome*.' ... then a bunch of genocide happened for 600 years, then Constantine co-opted Christianity. I'm not sure what point in there you want the robot to say that Rome's power was great. But since it said "all the people trying to get rome to stop being genocidal empirialists were the greatest senators", I have to say that it did a better job answering, than you did asking or recieving the question and answer. Again, i say this to help you help it grow, not to like, beat you up. This is just, the way that I know how to argue for growth. Love ya brother. :wizard:

[@soyhenryxyz]

wild, and has made me think of gpt's in a new frameworks almost as if we need to "divide" them to be more specific to specific tasks and have those agents communicate to come up with a proposed answer are there any downsides to putting all of these instructions into one prompt? great video!

[@Dsinatra3]

Wow I’ve been trying to put this into words and how to put it to chat gpt.

[@MarcoGillain]

I've JUST spent the last 12 hours doing exactly this! And I kid you not, exactly the same happened when you posted your AI doctor pipeline 😅 Lesson learned, I should go through your repos more thoroughly. Thank you for the incredibly valuable content Dave 🫡 Ps: Is it possible that the higher tier membership isn't available right now?

[@DaveShap]

$50 is still available

[@starblaiz1986]

This was the missing piece I needed for my own experiments with getting AI to automatically research and find accurate information, so thank you for diving deeper into this! ❤ What are your thoughts on filtering out misinformation and propaganda? I've been experimenting with having the AI evaluate sources over time and storing a quality score along with notable biases so that it can build up its own list and have some understanding on when it's appropriate to be skeptical, and when it's appropriate to trust. One of my biggest concerns is that the information space will soon be flooded with propaganda and misinformation mass-generated by AI's, and drown out voices of reason and rationality. Simply taking top results is an unstable approach, so the AI needs some mechanism to be able to say "hey, this may be a popular result, but it's actually not lining up with more trustworthy sources, so take it's information with some extra skepticism". What are your thoughts on how to approach this?

[@DaveShap]

You include metadata about the source eg name of the site or source, reputation, etc.

[@starblaiz1986]

@@DaveShap Yeh that's pretty much what I'm doing, and I'm storing that information in an SQL database that it can pull from and compare to while it's doing its search stage. I'm also finding having it identify logical fallacies is useful too, as typically poorer quality sources tend to step in more of them. Right now I'm just having it log an average number of logical fallacies per paragraph for each source, but I feel like the type and context of them is probably important metadata I should be capturing too, I just haven't figured out what to do with that information in the process yet.

[@stuartpatterson1617]

@@starblaiz1986 Interesting stuff! I would let it (BSHR) or another assistant evaluate and (update where necessary) the SQL DB as it uses the contained information in it's initial information search #MISSION query.

[@user-nz4hm1hu5u]

Maybe decentralization and blockchains will help with that, it's a major concern of mine as well.

[@Vaeldarg]

@@starblaiz1986 The problem with this is the sophistication of certain bad-actors that purposefully spread disinformation for state propaganda. China/Russia being the worst culprits, with China taking advantage of even the language barrier itself. The website for the company behind the "weather balloons" that were discovered flying over sensitive U.S military structures, had 2 versions of itself depending on what language viewed it in. The Chinese version was plastered all over with hammer-and-sickles, and boasted about these balloons being capable of carrying weapons of mass destruction....

[@8eck]

Would be cool to integrate some benchmarking solutions to see well overall solution is working, apart from reading all the outputs yourself. Good topic for another video i guess.

[@stumblinguponbliss]

How can we create/leverage a GPT out of this? Possible? And combine with your latest swarm insights?

[@MetricZero]

In layman's terms, what do I have to download, install, or provide, to create an AI assistant LLM that can take on a persona and remember all our conversations?

[@ethanmuhlestein8187]

I wonder if you could implement this idea for search augmented agents using autogen or something similar, and using the prompt from the previous video for search queries.

[@DaveShap]

Yeah the idea is that this will eventually be a plug and play component.

[@ethanmuhlestein8187]

Thanks! I appreciate the time and effort you put into your videos.

[@dewilton7712]

I tried to combine some of these ideas with Professor Synapse. I tried to have Professor Synapse use "thinking hats" as expert agents, but I would have to force it to use the "thinking hats" after the initial prompt or it would leave them out. After it starting using them they would not collaborate with each other or the conductor even after I told them too. I also was trying to have them and Professor Synapse use deductive reasoning and the Socratic method. I was also trying to have them all use information foraging and information literacy to come up with the questions. I do not know if I was stuffing the prompt too much and it was getting confused. How should I refine the prompt with these parameters?

[@godned74]

Ultimately you want to get to where you are using a server as well as an IDE to take as much weight off the llm as possible by storing your questions in the form of json's and custom instructions aswell as zip and or pictures as I do not believe it chews up any tokens. I was able to get free 7billion p llms to work almost as well as chat gpt and in some cases I can get it to chat indefinitely and with higher quality. I toot Professor synapse and transformed it into a master SPR prompt maker then took the prompts and insert some into json and others into the pythonic conversational template for the IDE. its a work in progress.

[@miblish5168]

I'm lost. I feel like I need to go away and study for a couple years and try again.

[@gregw322]

Ya dome looking shiny, bro. You must’ve buffed before recording. 🤣

[@mret36t]

Is technique 2 an easier plug n play/copy paste version?

[@thadgrace]

Is there a way to copy an entire conversation within ChatGPT? I noticed when I try to copy a conversation it only selects the block of text that I’m trying to copy instead of the entire conversation. I’m trying to do this from an iPhone if you have any answers to this, that would be great thanks.

[@ChaoticNeutralMatt]

Not within the apps. You could export the data though.

[@FinanceLogic]

Im not saying i know more than nasa or something but you're going to bias the data a lot not giving it the right longtail or shorttail keywords to start the base research with. And then brainstorm off that. Do another research and or retrieval step maybe after that. its like an automatic overfit on your preconceived notions. I would let it run free as early as you can steer it. (if and or when it's hooked up to the internet) paused at 5:29 Great Stuff so far. You always make me feel less frustrated about people taking nearly all my work for fre,e because you provide so much for so little. Thanks.

[@FinanceLogic]

I see around 13 minutes you explain that some of the searching is just skipped in mentioning because it's just a built in api call to a search engine and not what you're asking it to do directly, but i think it should be a part or parts of the loop to where it Can't just go to the easiest relevant thought, but also can't go as easily to where you accidentally pointed it to with our personal user biases.

[@Techtalk2030]

Not related to your video, but have you seen Disney’s bd-1 robot by any chance?

[@DaveShap]

I WANT ONE GIVE IT TO ME

[@Techtalk2030]

@@DaveShap I want one too. Hope they release them for sale lol

[@InAVanByTheRivr]

Hi david! I recently incorporated in canada and am in the process of filing a patent for a product in my industry. Im looking for a consultant who can help with deciding on which A. I. and how to integrate it into the business and product idea. Does your patreon typically help with these types of questions?

[@DaveShap]

There are plenty of other entrepreneurs, yes. But keep in mind it's not a formal consulting service. You can connect with similar folks

[@8eck]

Some simple examples under each section would be great, because almost all phases look too vague.

[@FrotLopOfficial]

Add Matt Wolfe in the title. He deserves it

[@paparaoveeragandham284]

look into 7

[@valdasdr8432]

Completely off-topic: wouldn't aliens be able to easily detect and hack our quantum computers? What if quantum field is just a cosmic internet for quantum mastered civilizations? Quantum computers solve Fermi paradox?

[@DaveShap]

No that's not remotely how QC works

[@abrasivepaste]

True, but it would make for a fun scifi book

[@mrNashmann]

Glad altmans gone go ilya

[@Dan-oj4iq]

In many of your videos over the past "years" have been about your interest in the Roman Empire and that time frame. I am curious as to why that period of time is of such interest to you. Whatever your reason, knowing you as little or as much as I do, your answer would help me realize what I have been missing in my own long life. Being one who finds that period of time to be of no interest to me at all. All I would know (guess) about the people of that time would be about their lack of physical hygiene. (Now of course I realize that is an extremely narrow point of view). At my time in life, I could use anything to hang onto.

[@DaveShap]

History nerd, and so is my wife. Plus Rome is cool.

[@Dan-oj4iq]

@@r3furbish3dbrain12 I actually wholeheartedly agree with you. It is simplistic to think of the Roman Empire as to its physical hygiene. But due to various things that is the first thing that comes to my mind. I guess societies that are able to create great things in life deserve better. It's just that.........

[@rasmusfoy]

@@DaveShap did you listen to the history of Rome podcast? Or Dan Carlin’s hardcore history episodes on the Roman Empire? I also liked James Clavell’s novels on Cesar. Highly dramatized and not perfectly historically accurate but his footnotes gives a good historical commentary.

[@Bakobiibizo]

You grossly underestimate the depths of depravity of the human race lol. "Not that I think anyone would. I dont think."

[@amandabeckner7162]

Thanks!

[@DaveShap]

Thank you!

[@lokiholland]

Thanks!

[@DaveShap]

Thank you!