Top AI channel! Not only this is the only RUclips video that properly tests the new models, but you also provide a million-dollar meta-prompt for free, no email required. You're a legend.
Bro your prompting blew my mind. Even if I didn't care about the o1 vs o1-Pro mode I am so glad I found this video. That flow chart is INSANE. Do you have some examples of meta prompts we can use? The tables are fantastic too. Man I can't wait to test this stuff out later this weekend.
I watch all of your videos. I do not know how to code yet, just starting, but have been a musician for my whole life. I wanna learn to do what you are doing!!! Been dreaming of functionality like this forever…very cool. Thank you for posting…even for us noobs, this gives us a shit ton of language and experience to help guide us forward. Hoping to catch up! Let me know if you ever want to learn music…would love to exchange lessons with a dude like you.
Hello! I’m a developer and at my workplace there are many musicians-turned-programmers. It’s very common! So keep putting in the effort and eventually you’ll get there! 🙂🙌🏻
Thank you! Can you share more of your library prompts you use for coding, for cursor setup , etc? Maybe on github open source it? (If possible) Also, please do a visual on your curser config and setup. I seen you use some key shortcuts to generate parts of your meta prompt which is a time saver. How did you set it up? Will be a great video on your complete cursor workflow.
I'm just curious: do you have any examples of actual real-world work getting done with these methods? Real world SUCCESSFUL end products? I'm loving this content, but sometimes it feels like a lot of bells & whistles without showing how it's changing your workflow in a meaningful way where you are dealing with problems that arise day-to-day when building software for paying clients. I'd love to see some content focused on that value proposition of adopting these methods and integrating these tools.
I don't want to be rude. Really, I don't. I am very sorry if you take this the wrong way. But here is the thing: How the hell, how the hell does someone who knows a little bit and the tech / coding business NOT SEE the HUGE advantages of LLMs for coding. How? Every day for the last year and a half at least, I wake up with an idea and code it up in MAX a few hours (and I suck at coding btw) with Cursor. I went from a very average coder to pumping out a crazy amount of actual products that I could sell (some with great successes). If you ever coded, or tried to sold anything online. You KNOW how powerful these "methods" are. I hope you don't take it the wrong end buddy. Cheers
@@vincentjean6756 Perhaps I was unclear; I use LLM tools daily (Cursor, SuperMaven, CoPilot, Windsurf, etc... tried them all). They have absolutely improved my workflow. What I am asking is how THESE methods, meta prompting and chaining models, really moves the needle. Day-to-day work doesn't really involve the need for these seemingly over-engineered methods, unless I am missing something...hence, why I asked. 🙂
Differences seem to be marginal while the prices are vide-versa. it would be interesting to see what other really complex problems can the Pro Mode ChatGpt can solve? As you have mentioned, one needs to understand and explore this part...
Remember that $20/month only includes 50 prompts per week. While the difference between pro mode and non-pro mode o1 may be considered marginal by some, 50 prompts per week is miniscule, if you are using AI everyday to transform a business the difference between 50 prompts and unlimited is not.
@@nicolasivorranihe just explained you why it is worth it. It s as if saying 200 USD/month for unlimited fuel ⛽ is not worth it. You just don't make any sense
These videos are so good. This is the best channel IMO about AI for engineers. So thank you for producing such great content and providing the prompts for us to try it out on gist. Are you planning on releasing something about Anthropic's MCP and how this can impact how we use tools?
hi! How can I expand the output of an llm? Currently it has a small output tokens limit. I would like to generate more extensive answers' Is there any technique tu cut off the answers and generate more than one answer? For example, if I ask the AI to create a book, how can I told the ai in the prompt to create multiple responses (and each chapter mantains a consistency with the rest of the chapters
I agree with your accessment that if you are a power/ business user *that's working in a team,* then yes the $200 a month is worth it. Emphasis on TEAM because it's possible to have 2+ users log in to the same chat GPT at the same time right now. I imagine pro will be no different. So if you get 5 people on the same account that's $40 each. But if you're just 1 guy, then I would just stick with the basic plan. Also I'm late to the meta prompting game, but I just realized that you can use meta prompting to turn LLM into a mini agent swarm by divying up tasks and tools per task. If only o1 had access to custom GPTs as well :(
Even more confused that you fed it in API prompt code instead of uploading a JSON file and just asking it to summarize it for you, giving you top 5 positive and top 5 negative with highlights on key topics.
No, I do not behave differently based on whether instructions are framed as plain English or presented in a structured format like XML, JSON, or other markup-like styles. My behavior is determined by the content and intent of the input, not its formatting. If the provided input, whether in plain text or structured format, contains a clear request or instruction, I’ll interpret and respond accordingly. The example you provided (e.g., with and tags) would simply be parsed for its meaningful content, just as if it were phrased in natural language without tags. The tags themselves do not influence my behavior unless explicitly programmed to do so, which is not the case here. Feel free to use whatever format is comfortable or best for structuring your request!
Meanwhile, in other room, engineers use 8B models, to remove hooks from blog articles, and extract bits of worthy ideas from tons of auto-generated content.
Thanks for the prompt structure, I'll do some tests to see if the output is improved and compare with my current prompt structure. what do you use for transparency on macOS?
Great video!❤😊 “I’m looking for an AI tool that can quickly translate RUclips videos into Danish and play them with Danish speech. Ideally, it should be able to extract subtitles, translate them, and then use text-to-speech technology to provide the audio in Danish, all in sync with the video. Any recommendations or solutions would be greatly appreciated!”
Hi Dan, do you do any consulting. I am working on building an engine to help startup founders and it is clear to me the value of the prompting structure. Can you let me know if I can get in touch. Thanks Dave
I accidentally fell into learning how to be a prompt engineer/ product engineer after starting my fitness coach company…. And am shocked at its capacity. I’m really good at it too. How do I find a job that integrates this skill without having a degree in computer science? Are there any discord servers for people like us?
Thank you @indydevdan. Great comparison. o1 pro nit for the average Joe. I wish I was a software engineer. Someone could make a channel that breaks down your work for us newbies 😂 I would love Ava on my pcas well!
It seems like a good workflow could be having 01 pro mode take on the role of a high level engineer in an agent system and then something like Claude doing the coding. It could even be possible to just hack that together with cursor by using pro mode to do the high level thinking and then copying that into cursor for it to execute the plan
What is the longest prompt that you can input? How long of a volume can you use? Did you try to input a 50.000 word novel in it? Did you consider what the user load does to the AI? You did not even come close to test this AI.
Hi Dan. Just letting you know that I could not like your video because YT did not present a thumbs-up like button ... only a dislike ... so this might be why you have a bunch of dislikes ... I mean what's not to like right? YT has had this bug on a couple of videos I watched recently, so maybe this is something you can get them to address. Consider this comment a like!
Hey Dan love your content but I find myself ejecting lately after 3-min because it’s not clear to me what the problem is and what you are hoping to demonstrate so I can gauge my interest on whether or not to continue the 30-min journey with you
That's what I was wondering, as well. Natural language IS the interface with LLMs. Layering on XML seems like a backwards abstraction that doesn't really gain a lot of clarity and accuracy, in my experience. There's nothing in any LLM documentation that states it requires structured inputs to provide structured outputs. That's part of the draw.
@@larsfaye292XML can be incredibly powerful in prompt engineering. It is one of the best ways to keep the llm on track when giving complex instructions. My complex reasoning prompts fall apart when xml tags are omitted.
Top AI channel! Not only this is the only RUclips video that properly tests the new models, but you also provide a million-dollar meta-prompt for free, no email required. You're a legend.
you created a video and same time got your 40 minutes of hand workout in, good job.
😂😂😅😊
About a weeks' worth in 40 minutes! LOL
he has nice hands
@@thunkin-ai Its AI, its not really his hands... #Mindblown
Bro your prompting blew my mind. Even if I didn't care about the o1 vs o1-Pro mode I am so glad I found this video. That flow chart is INSANE. Do you have some examples of meta prompts we can use? The tables are fantastic too. Man I can't wait to test this stuff out later this weekend.
What extension are you using that has the token count on the bottom?
Token Count in Status Bar for your active file and selection
Token Count by Marxism
I watch all of your videos. I do not know how to code yet, just starting, but have been a musician for my whole life. I wanna learn to do what you are doing!!! Been dreaming of functionality like this forever…very cool. Thank you for posting…even for us noobs, this gives us a shit ton of language and experience to help guide us forward.
Hoping to catch up! Let me know if you ever want to learn music…would love to exchange lessons with a dude like you.
Hello! I’m a developer and at my workplace there are many musicians-turned-programmers. It’s very common! So keep putting in the effort and eventually you’ll get there! 🙂🙌🏻
@ Thank you! 🙏
I don't speak a word of English, but thank you for the quality of the content available here and free of charge. 🇫🇷
finally someone that made a real challenge for o1pro
Thank you! Can you share more of your library prompts you use for coding, for cursor setup , etc?
Maybe on github open source it? (If possible)
Also, please do a visual on your curser config and setup. I seen you use some key shortcuts to generate parts of your meta prompt which is a time saver. How did you set it up?
Will be a great video on your complete cursor workflow.
Next level prompting analysis. Awesome work
Love your work! Maybe the next level is to automate the craft of agents with the prompt created by the metaprompt and the code generated also?
what extension are you using for the token count?
10:45 OMG - I see where you’re heading with this! Wow, man, next level thinking.
I'm just curious: do you have any examples of actual real-world work getting done with these methods? Real world SUCCESSFUL end products? I'm loving this content, but sometimes it feels like a lot of bells & whistles without showing how it's changing your workflow in a meaningful way where you are dealing with problems that arise day-to-day when building software for paying clients. I'd love to see some content focused on that value proposition of adopting these methods and integrating these tools.
can only second that. Exactly the same thoughts while my brain hurts after watching this for 15 minutes :)
I don't want to be rude. Really, I don't. I am very sorry if you take this the wrong way. But here is the thing:
How the hell, how the hell does someone who knows a little bit and the tech / coding business NOT SEE the HUGE advantages of LLMs for coding. How? Every day for the last year and a half at least, I wake up with an idea and code it up in MAX a few hours (and I suck at coding btw) with Cursor.
I went from a very average coder to pumping out a crazy amount of actual products that I could sell (some with great successes).
If you ever coded, or tried to sold anything online. You KNOW how powerful these "methods" are.
I hope you don't take it the wrong end buddy.
Cheers
@@vincentjean6756 Perhaps I was unclear; I use LLM tools daily (Cursor, SuperMaven, CoPilot, Windsurf, etc... tried them all). They have absolutely improved my workflow. What I am asking is how THESE methods, meta prompting and chaining models, really moves the needle. Day-to-day work doesn't really involve the need for these seemingly over-engineered methods, unless I am missing something...hence, why I asked. 🙂
@@vincentjean6756 based response
PydanticAI is probably more what you are looking for then
Differences seem to be marginal while the prices are vide-versa. it would be interesting to see what other really complex problems can the Pro Mode ChatGpt can solve? As you have mentioned, one needs to understand and explore this part...
Loving your work. Thank you for the brilliant insights. Count me in on the course when youve finished compiling.
When you said "hands on" you weren't kidding!!
Remember that $20/month only includes 50 prompts per week. While the difference between pro mode and non-pro mode o1 may be considered marginal by some, 50 prompts per week is miniscule, if you are using AI everyday to transform a business the difference between 50 prompts and unlimited is not.
It's 200 dollars 😅
There are new models everyday, 200$ is just not worth it
It gives you 50 prompts of “amateur” o1 mode. For even one “pro” mode prompt, you need to lay down 200
@@nicolasivorranihe just explained you why it is worth it. It s as if saying 200 USD/month for unlimited fuel ⛽ is not worth it. You just don't make any sense
@@stevenfullman5646 I think 200 dollars is way too much now after some time some api service will give it in a cheaper cost.
These videos are so good. This is the best channel IMO about AI for engineers. So thank you for producing such great content and providing the prompts for us to try it out on gist.
Are you planning on releasing something about Anthropic's MCP and how this can impact how we use tools?
Could you share more context about aloe1 command at 27:38?
Amazing! This is awesome! Let's push!!!
I'd love to see the meta promoting is used to develop a complete app
Great! Waiting for AI COding Course
Can tou guide me to where is the best course i can take for promot engineering?
hi!
How can I expand the output of an llm? Currently it has a small output tokens limit. I would like to generate more extensive answers'
Is there any technique tu cut off the answers and generate more than one answer?
For example, if I ask the AI to create a book, how can I told the ai in the prompt to create multiple responses (and each chapter mantains a consistency with the rest of the chapters
Second watch changing notifs to ALL GREAT channel and detailed clean spoken breakdown with value not garbled word salad and hype
How can this be applied to trading or stock charts?
I was curious about benchi, but I don't see any link about this tool. Thank you very much!
Confession time: I watch Dan at 1x speed 👀
Don’t brag about it! 😂
I agree with your accessment that if you are a power/ business user *that's working in a team,* then yes the $200 a month is worth it.
Emphasis on TEAM because it's possible to have 2+ users log in to the same chat GPT at the same time right now. I imagine pro will be no different.
So if you get 5 people on the same account that's $40 each. But if you're just 1 guy, then I would just stick with the basic plan.
Also I'm late to the meta prompting game, but I just realized that you can use meta prompting to turn LLM into a mini agent swarm by divying up tasks and tools per task. If only o1 had access to custom GPTs as well :(
the thumbnail is interesting!
Hi. What code editor are you using here?
How do you make cursor or any other text editor show token count in the lower right corner?
token count extension by marxism
What software is used for your mouse cursor for screen capturing has the rounded aura around the mouse?
cursor pro
I’m a little confused. These prompts are guide-rails for making specific types of requests as opposed to just chatting with a LLM?
Even more confused that you fed it in API prompt code instead of uploading a JSON file and just asking it to summarize it for you, giving you top 5 positive and top 5 negative with highlights on key topics.
No, I do not behave differently based on whether instructions are framed as plain English or presented in a structured format like XML, JSON, or other markup-like styles. My behavior is determined by the content and intent of the input, not its formatting.
If the provided input, whether in plain text or structured format, contains a clear request or instruction, I’ll interpret and respond accordingly. The example you provided (e.g., with and tags) would simply be parsed for its meaningful content, just as if it were phrased in natural language without tags. The tags themselves do not influence my behavior unless explicitly programmed to do so, which is not the case here.
Feel free to use whatever format is comfortable or best for structuring your request!
Meanwhile, in other room, engineers use 8B models, to remove hooks from blog articles, and extract bits of worthy ideas from tons of auto-generated content.
I couldn't find a tool called "aloe1", is the same as "aloe" on Github ?
Thanks for the prompt structure, I'll do some tests to see if the output is improved and compare with my current prompt structure.
what do you use for transparency on macOS?
edit the screen layer by dropping opacity in post
Great video!❤😊
“I’m looking for an AI tool that can quickly translate RUclips videos into Danish and play them with Danish speech. Ideally, it should be able to extract subtitles, translate them, and then use text-to-speech technology to provide the audio in Danish, all in sync with the video. Any recommendations or solutions would be greatly appreciated!”
Thanks for the top tier content
Hi Dan, do you do any consulting. I am working on building an engine to help startup founders and it is clear to me the value of the prompting structure. Can you let me know if I can get in touch. Thanks Dave
Awesome video 😊
commenting for visibility and to thank you for the value i get out of it
Thank you! appreciate your knowledge!
Sorry for asking a stupid question. I just subscribed pro ; how do I get access to ChatGPT o1 Pro mode? I can't find it anywhere.
Left corner
One major component of making this decision is how often are you using it? AFAIK O1 pro is unlimited while plus has that 50 messages/week limit.
Trying to find use cases for iOS dev via meta prompting
It’s much simpler: you need o1pro when you run out of quota on o1 or advanced audio chats
I accidentally fell into learning how to be a prompt engineer/ product engineer after starting my fitness coach company…. And am shocked at its capacity. I’m really good at it too. How do I find a job that integrates this skill without having a degree in computer science? Are there any discord servers for people like us?
Thank you @indydevdan. Great comparison. o1 pro nit for the average Joe. I wish I was a software engineer. Someone could make a channel that breaks down your work for us newbies 😂 I would love Ava on my pcas well!
I still want finger puppets on your hands
Great videos tho
Great content it’s soooo interesting
It seems like a good workflow could be having 01 pro mode take on the role of a high level engineer in an agent system and then something like Claude doing the coding. It could even be possible to just hack that together with cursor by using pro mode to do the high level thinking and then copying that into cursor for it to execute the plan
Fantastic Video! Thank you.
I also find studying philosophy is useful for coming at the AI from unique directions.
I like how you filmed your hands gesturing over your computer for a long time. lol
I like this too :) Faceless videos with _some_ kind of exposure (even hands) feels more real than totally faceless and automated.
Bro is playing chess and we are playing checkers.
Whats funny is "Meta" is quite literally "higher dimension"
Background is so much distractive, please make background just solid, really man, your hands are moving and making hard to focus
MetaPrompting !! I am all about that Meta stuff
awsome value bro
why is this xml and not json schema?
1:07 "hands on examples..."
Meta prompting is simply structured input in xml 🤯
I tried using the meta prompt and it did not respond with what you got haha. I think I am understanding how to use it wrong.
In every chat, there is a copy option that will copy the content in mermaid format
What is the longest prompt that you can input? How long of a volume can you use? Did you try to input a 50.000 word novel in it? Did you consider what the user load does to the AI? You did not even come close to test this AI.
Generaly O1 pro thing 1-2 minutes, do i ask to much?
If pro mode had better abstract or wide thinking id be more than happy to pay for it. For now ill pay $20/month for the chain of thought
Amazing that this is free
Hi Dan. Just letting you know that I could not like your video because YT did not present a thumbs-up like button ... only a dislike ... so this might be why you have a bunch of dislikes ... I mean what's not to like right? YT has had this bug on a couple of videos I watched recently, so maybe this is something you can get them to address. Consider this comment a like!
"Could you please add Turkish subtitles? I really enjoy your videos, but I can't fully understand them due to the language barrier. Thank you!"
Hey Dan love your content but I find myself ejecting lately after 3-min because it’s not clear to me what the problem is and what you are hoping to demonstrate so I can gauge my interest on whether or not to continue the 30-min journey with you
Is there some TLDR on this? 😥😥
The moment of truth: 13:57
the excited subtext tone throughout gets a bit monotonous. you need your AI to vary the tone a bit. maybe that can go in your prompt?
Personally, I don't see the difference in output that's worth 200USD, but who has money to spend why not...
Great content! thanks 😊 what is the Algolia link you are using at ruclips.net/video/yZGb9-Z9DG0/видео.html can't figure it out
I did too.
iDK about xml. Seems like an unnecessary PITA with being able to type LLMs.
That's what I was wondering, as well. Natural language IS the interface with LLMs. Layering on XML seems like a backwards abstraction that doesn't really gain a lot of clarity and accuracy, in my experience. There's nothing in any LLM documentation that states it requires structured inputs to provide structured outputs. That's part of the draw.
@@larsfaye292XML can be incredibly powerful in prompt engineering. It is one of the best ways to keep the llm on track when giving complex instructions.
My complex reasoning prompts fall apart when xml tags are omitted.
@@larsfaye292 Actually, it's the recommended format by Anthropic, and its effectivity was thoroughly tested in another video in the channel.
All the replies are missing for me on this video
So true so true
I don’t understand
Me either and I'm trying to identify the prerequisite knowledge I need to use meta prompting like this.
@ he has the coolest website ever. Enter de matrix den
33:33 so true
Can you teach me?
i hope these reasoning models don't become free
Smart
I brought the 1000'th like lol
you just nuked pro mod
Omg
Sorry this was just a waste of time
which tool are you using as a text editor? it is looking too slick 🧊