Wow, Professor Synapse, your custom prompt for ChatGPT is a game-changer! The clever approach you've taken has not only made my interaction with AI incredibly efficient but also so much more engaging. Thank you for dedicating your time and effort to create this; it's clear you've put a lot of thought into it. 💪
This is brilliant, I just learned the beginnings of basic logic, and, or, ect..., I mean learned not told me. it help me figure out how a lot of it works. What would always get me lost was converse part of it. I took a break to think about what it was saying came back to it and asked it if it was X(not going to say because ones own take on how to get something is unique). It made the light bulb go off in me. I asked to integrated the learning based off of two books Make it Stick and A Mind for Numbers. It has been a, wonderful, mind opening experience. I never thought, I could do this, every time I tried i would get lost when people explained i was more confused then after their answer. I tried repeatedly over 20 years. I always felt like there was something missing or I missed something that I should of had before starting. There was and with your prompt GPT actually help me figure out that missing part. I tried doing this without your prompt. I kind of understood what it was saying, however, I didn't learn. It is in my head now and the possibilities of actually understand discrete math, and help me understand logic while learning python seem more realistic than before. Thank you for this 🤯😁
The prompt you have provided has significantly improved the quality of all GPT responses in general, Thank you. I started modifying it to be even more observant of user goals(those might not be obvious - even to the user, or at least not well verbalized) and to ask questions - but only when necessary.
That's awesome. I saw your interview on the Goda Go channel. At first I thought I had to change the descriptors in curly brackets, but this video made it very clear. Can't wait to test it out in planning and action plans for various online projects, like TH-cam, publishing colouring books, print on demand, and some affiliate marketing. Time to buckle up and get this tech working for me now.
Stumbled upon this today, and I have to say, I'm very impressed! (And I don't impress easily :D) Thanks for sharing this and helping us, the AI community, learn this technology as it evolves along side us.
You sir, deserve to go to the Moon 🚀🚀 I can't wait to see what we all do with this! I have it helping me setup automated Ai fueled income streams right this moment!
This stuff is so darn cool. I wish I were savvy enough to keep up with it all. I'm trying to use this tech to create a workflow that will either know or prompt me to upload my professional xp, target job description, then, output a tailored resume to suit. I'll be working on this so if anyone has some pointers, send 'em my way!
Slightly concerned by the hallucinations that will be caused from the token window/limit per the length of this. It’s memory is/should be quite a bit shorter when attempting to get accurate results. Especially for zero shots. It exponentially decreases its gradient weighting for ongoing context. Basically new messages will be forced to replace something at all, and it’s going to pick the thing that has not been promoted yet to ensure proper scoring in sumilarity matches. Chaining works because it maintains clean token reserves. Using all of that up right away even for chaining will wrangle everything.
Have you tried it? I've actually gone WAY beyond the token limit, especially when it's in the custom instructions. Similarly the /save command anecdotally has worked very well in helping to maintain the context.
I like this a lot as well my only issue is that for a large project it keeps jumping from new agent to new agent without finishing the initial requests and over time I just get sent to like the 20th different agent and I haven't gotten anything accomplished yet. Is there any way to get it to remember everything that we've discussed from the beginning and tie all of the different agents together into one cohesive project?
Trying out this prompt, fairly interesting so far, thank you! It does not have browsing capability, at the moment? Is there any way to add this or access a plugin in addition to Code Interpreter?
You cannot use code interpreter and plugins at the same time, but there are a couple web browsing plugins such as webpilot and keymate.ai. Theyre pretty janky though.
@synapticlabs Hi Joseph! I've been using your Synapse_COR prompt/system in my Custom Instructions and I REALLY love it overall, but there's a minor issue I have with the system that I need your input on. I noticed that Synapse_COR always asks me whether or not it should proceed with the next step in between every single step before the completion of the goal at hand. This is an issue because as you know GPT-4 has a 50 message cap per every 3-hour period. When the Synapse_COR system asks those unnecessary “shall I proceed” questions, it quickly burns through my GPT-4 message "budget". I have tried to slightly modify the Custom instructions to force Synapse_COR to meet the goal without asking whether to proceed every step of the way, but it's not working. Do you have a way of making Synapse_COR asks questions only AFTER it has met the goal I have assigned to it?
Hey there happy you’re finding value. Just something to note that ChatGPT can only output about 500 words max per message anyway so whether or not it asks you’ll likely have the same issue. You can always change the rules section to your hearts content though by experimenting with things like “always output the most text you can” or something. That being said the ask is important since it allows you to realign and make sure it’s not wasting your messages by doing something you don’t want it to do.
You could save yourself some steps by giving it as much info as possible so it doesn't have to ask you some questions. Like in the example in this video, he could have stated in one of the messages that the main character was a Fox and that he wanted to the Professor to come up with everything. That way it wouldn't have had to ask him and he could save a prompt or two. Don't make it ask about what you already can provide so to speak.
Create prompt! I added this to rules: "-Focus only on practical, ChatGPT web-interface tasks. Avoid theory and averages. Provide real specific data, reason step-by-step for precise numbers."
I started using this amazing prompt (thank you a lot for it!!!!). The behavior of the Chat-GPT has changed a lot. She began to move more actively towards the implementation of the task. But I didn't enter the /start command. Does it mean that the prompt is not working properly? I started writing a story with his help. During the dialogue, Professor Synapse did not call his experts. And after discussing the details of the plot and characters, Professor Synapse said that he would write the story in 1-2 weeks. Is it normal?
It should still work if you don’t do start, but that just makes sure you get it started on the right foot. For the 2 weeks it’s definitely hallucinating. What you see if what you get in chatgpt (for now). It’s playing its role too hard and saying what an employee might say, but it’s not actually going and doing it. What model are you using? If 3.5 I would recommend using the custom instructions version, the full version is too much for 3.5.
@@synapticlabs I'm using version 4. But I've only been working with chatGPT for a week and don't know all the basic rules yet. I still don't understand what user instructions are. I realize that I am still very new to all this. But I have a great desire to understand it and use it to create different projects.
It’s under the MIT license. Feel free to take and make derivations but if you use it commercially don’t call it Professor synapse. Any recognition would be greatly appreciated directed toward Synaptic Labs
I dont know if my question is sensible or not but they say if you don't ask you wont know. So my question is: Do I need to edit the prompt or I can just paste it the way its and let it do is thing? Thanks for this and your your contributions to the Ai world.
It will be less likely to work with something like Claude or Pi, since those are trained to have their own personality essentially, but should work on other models.
Objectives for what? It works for any sentence transformer actually even Claude. It’s just that those models are built to fine tune and fit to tune. Their purposefully smaller token limits make accuracy much easier to develop as they’re not clouded with overly generic content from every human on the planet.
It is admittedly untested experimentally but the idea is that if you are summarizing and stating the next step consistently it’s less likely to lose the thread of the conversation when it hits the context window. Anecdotally I’ve been able to extend a conversation well past the window and it reliably maintains its memory
before trusting it to help me with my research, i challenged it to correctly present me with the prime factors of 7877. surely, it can do such a simple thing; surely, i'm only misusing the prompt. please make a video to explain how to get it to correctly answer something like that.
If you're trying to use LLMs for math, you're using a screw driver to hammer a nail. LLMs aren't good at math. Similarly research is not a good use of it unless you are feeding it the research you want to use as a basis, or maybe to help brainstorm some research questions, or support you in writing up the research based on your notes. LLMs are a type of hyper advanced autocomplete based on their training data. You can get around this outside of chatgpt by connecting an API to a calculator, or a research database (you can use the arxiv plugin and scholarAI if you want), but just by itself it's going to get things wrong and hallucinate.
this prompt does not work correctly in Portuguese. I tried to use it with commands in Portuguese, he understands, changes the language, but does not start the intelligent agents to help. Do you know how I can do to solve this?
ChatGPT isn't very good at any other language than english...I imagine you already translated the entire prompt into portuguese? You might want to add a rule that says "only respond in portuguese" or something...in portuguese
@@synapticlabs i translated, he responds in portuguese, the problem is that it doesn't perform the specific functions, it doesn't call another customized assistant for each task. I tried it on chatgpt 4 and it worked, however, I wanted to try it on 3.5, the turbo is faster
I cant understand how people are still so low level at prompting. Like that prompt idea is many months old, we are ages ahead of that. How people don't know the real stuff?
The clarifying questions are annoying because you spend all this time clarifying and then response is some generic non-sense that wastes the last 5 minutes of your interaction. Why does almost every interaction require a clarification. Its not clear what the clarification even does in terms of changing the answer if anything. It just seems like more interactive prompts that don't necessarily produce better results.
Sorry it’s not super useful to you but highly recommend you make some of your own rules to make it respond closer to the way you like. Tell it to ask fewer questions as well.
Wow, Professor Synapse, your custom prompt for ChatGPT is a game-changer! The clever approach you've taken has not only made my interaction with AI incredibly efficient but also so much more engaging. Thank you for dedicating your time and effort to create this; it's clear you've put a lot of thought into it. 💪
Glad it was helpful!
Was this comment written with ChatGPT?
@@_arman_ 🤣
Thank you for making this and sharing it with the world. Salute!
Thank you for deciding to share, we will all benefit from your generosity.
Brilliant. Had to find source because so many taking your shine by copying your video and reviewing it. Keep up the good work.
Cool prompt, I just tried that prompt for a Google ad campaign.
It worked really well and wrote a really good ad.
Amazing! Thanks for your generosity in sharing.
My pleasure!
This is brilliant, I just learned the beginnings of basic logic, and, or, ect..., I mean learned not told me. it help me figure out how a lot of it works. What would always get me lost was converse part of it. I took a break to think about what it was saying came back to it and asked it if it was X(not going to say because ones own take on how to get something is unique). It made the light bulb go off in me. I asked to integrated the learning based off of two books Make it Stick and A Mind for Numbers. It has been a, wonderful, mind opening experience. I never thought, I could do this, every time I tried i would get lost when people explained i was more confused then after their answer. I tried repeatedly over 20 years. I always felt like there was something missing or I missed something that I should of had before starting. There was and with your prompt GPT actually help me figure out that missing part. I tried doing this without your prompt. I kind of understood what it was saying, however, I didn't learn. It is in my head now and the possibilities of actually understand discrete math, and help me understand logic while learning python seem more realistic than before.
Thank you for this
🤯😁
The prompt you have provided has significantly improved the quality of all GPT responses in general, Thank you.
I started modifying it to be even more observant of user goals(those might not be obvious - even to the user, or at least not well verbalized) and to ask questions - but only when necessary.
would you mind sharing your improved version? please 🙏 🥺🥺
I used this prompt for the first. Wow fantastic!!
Well done mate! Wonderful mind you've got!
Thank you kindly!
That's awesome. I saw your interview on the Goda Go channel. At first I thought I had to change the descriptors in curly brackets, but this video made it very clear. Can't wait to test it out in planning and action plans for various online projects, like TH-cam, publishing colouring books, print on demand, and some affiliate marketing. Time to buckle up and get this tech working for me now.
You got this!
Will this also work with Bard, Claude or Perplexity too?
Wow. Excellent work. You get a thumbs up for this. Brilliant prompt.
Stumbled upon this today, and I have to say, I'm very impressed! (And I don't impress easily :D) Thanks for sharing this and helping us, the AI community, learn this technology as it evolves along side us.
Wow, thank you!
Amazing work. Thank u for sharing
Great job!!! Thanks for sharing!! It takes AI assistants to the next level!!! Thanks!
Make sure you check out super synapse! This video is a few generations old at this point.
Super cool. Just starting to use it, but I can tell; it's going to be amazing. Thanks for sharing.
Thank you so much for this super prompt. It has been great to work with it - many thanks🍻
It's fun to work with, and easy to set up, thank you!!!
You are so welcome!
This is excellent. I think by far the /save feature is what sets this apart. The token limit has always been a problem for me. Thanks
Thank you very much for sharing!
You sir, deserve to go to the Moon 🚀🚀 I can't wait to see what we all do with this! I have it helping me setup automated Ai fueled income streams right this moment!
Much appreciated!
This is amazing!! Thanks for sharing :))
Amazing!!
thanks much sir for sharing this great info.👍👍
My pleasure
Awesome👌
Thank you for this.
Your welcome, and spread it around!
I just saw this prompt on another TH-cam video.
This guy invented it
Wonderfull! It's really a gamer changing spot!!! Thx
🤩🤩🤩🤩🤩🤩
Ths is a great prompt. How often should the /save option be used?
I usually do it after each sort of sub task or when I feel like I’m not clear on a next step
🎯 Key Takeaways for quick navigation:
Made with HARPA AI
This stuff is so darn cool. I wish I were savvy enough to keep up with it all. I'm trying to use this tech to create a workflow that will either know or prompt me to upload my professional xp, target job description, then, output a tailored resume to suit. I'll be working on this so if anyone has some pointers, send 'em my way!
I hope the professor can help you! Use the prompt to explain what you need and it should do the trick.
Thanks a lot. Could you alter it for it to work with Bing?
You can try, but probably not as well since Bing's chatbot has its own system prompt which might interfere. Experiment, though!
Slightly concerned by the hallucinations that will be caused from the token window/limit per the length of this. It’s memory is/should be quite a bit shorter when attempting to get accurate results. Especially for zero shots.
It exponentially decreases its gradient weighting for ongoing context. Basically new messages will be forced to replace something at all, and it’s going to pick the thing that has not been promoted yet to ensure proper scoring in sumilarity matches.
Chaining works because it maintains clean token reserves. Using all of that up right away even for chaining will wrangle everything.
Have you tried it? I've actually gone WAY beyond the token limit, especially when it's in the custom instructions. Similarly the /save command anecdotally has worked very well in helping to maintain the context.
Nice ,lets have timelines in your next videos
I’m a noob thank you for the feedback!
I like this a lot as well my only issue is that for a large project it keeps jumping from new agent to new agent without finishing the initial requests and over time I just get sent to like the 20th different agent and I haven't gotten anything accomplished yet. Is there any way to get it to remember everything that we've discussed from the beginning and tie all of the different agents together into one cohesive project?
which version are you using (custom instructions vs GPT)? and are you on GPT4?
Hi, I'm note sure why I copy and paste the prompt.txt to my custom instruction box, the professor icon turn into xxx
You don’t need it or you can replace it with your own. The prompt is also on the description
Trying out this prompt, fairly interesting so far, thank you! It does not have browsing capability, at the moment? Is there any way to add this or access a plugin in addition to Code Interpreter?
You cannot use code interpreter and plugins at the same time, but there are a couple web browsing plugins such as webpilot and keymate.ai. Theyre pretty janky though.
hope you see this... What plugins do you recommend to use with this amazing prompt?
I honestly find plug-ins pretty unreliable and now that web browsing is back I would just use that.
Is the still the Last prompt or have you come up with a improved version?
I continually update this one so it’s changed a little since this video. Then there’s super synapse which I have a video on as well you can check out.
@synapticlabs Hi Joseph! I've been using your Synapse_COR prompt/system in my Custom Instructions and I REALLY love it overall, but there's a minor issue I have with the system that I need your input on. I noticed that Synapse_COR always asks me whether or not it should proceed with the next step in between every single step before the completion of the goal at hand. This is an issue because as you know GPT-4 has a 50 message cap per every 3-hour period. When the Synapse_COR system asks those unnecessary “shall I proceed” questions, it quickly burns through my GPT-4 message "budget". I have tried to slightly modify the Custom instructions to force Synapse_COR to meet the goal without asking whether to proceed every step of the way, but it's not working. Do you have a way of making Synapse_COR asks questions only AFTER it has met the goal I have assigned to it?
Hey there happy you’re finding value. Just something to note that ChatGPT can only output about 500 words max per message anyway so whether or not it asks you’ll likely have the same issue. You can always change the rules section to your hearts content though by experimenting with things like “always output the most text you can” or something.
That being said the ask is important since it allows you to realign and make sure it’s not wasting your messages by doing something you don’t want it to do.
You could save yourself some steps by giving it as much info as possible so it doesn't have to ask you some questions. Like in the example in this video, he could have stated in one of the messages that the main character was a Fox and that he wanted to the Professor to come up with everything. That way it wouldn't have had to ask him and he could save a prompt or two. Don't make it ask about what you already can provide so to speak.
Create prompt! I added this to rules: "-Focus only on practical, ChatGPT web-interface tasks. Avoid theory and averages. Provide real specific data, reason step-by-step for precise numbers."
Be careful of hallucinations! Always evaluate yourself to make sure what it’s saying is true.
How do using different completion dates change things?
I started using this amazing prompt (thank you a lot for it!!!!). The behavior of the Chat-GPT has changed a lot. She began to move more actively towards the implementation of the task. But I didn't enter the /start command. Does it mean that the prompt is not working properly? I started writing a story with his help. During the dialogue, Professor Synapse did not call his experts. And after discussing the details of the plot and characters, Professor Synapse said that he would write the story in 1-2 weeks. Is it normal?
It should still work if you don’t do start, but that just makes sure you get it started on the right foot. For the 2 weeks it’s definitely hallucinating. What you see if what you get in chatgpt (for now). It’s playing its role too hard and saying what an employee might say, but it’s not actually going and doing it. What model are you using? If 3.5 I would recommend using the custom instructions version, the full version is too much for 3.5.
@@synapticlabs I'm using version 4. But I've only been working with chatGPT for a week and don't know all the basic rules yet. I still don't understand what user instructions are.
I realize that I am still very new to all this. But I have a great desire to understand it and use it to create different projects.
@@synapticlabs Do you think that chatGPT should write a story immediately, in a dialog box, during the current session?
@@ВалерийВовк-ш8ц chatgpt can only output so much at a time so you need to chunk things out in pieces. User instructions are the prompt you give it.
It's there any license on this if we create derivatives? If we modify and adapt it for a GPT do we need to credit you? If so, how?
It’s under the MIT license. Feel free to take and make derivations but if you use it commercially don’t call it Professor synapse. Any recognition would be greatly appreciated directed toward Synaptic Labs
@@synapticlabs Understood. Thanks!
I dont know if my question is sensible or not but they say if you don't ask you wont know. So my question is:
Do I need to edit the prompt or I can just paste it the way its and let it do is thing?
Thanks for this and your your contributions to the Ai world.
simply copy/paste the entire prompt, and input "/start" - you can edit the rules and commands section though to your hearts content.
Is this prompt compatible with all conversational AI? Does this prompt meet all the objectives?
It will be less likely to work with something like Claude or Pi, since those are trained to have their own personality essentially, but should work on other models.
@@synapticlabs doesn't work with Bing, Bing will not cooperate :) "It might be time to move onto a new topic. Let's start over."
Objectives for what? It works for any sentence transformer actually even Claude. It’s just that those models are built to fine tune and fit to tune. Their purposefully smaller token limits make accuracy much easier to develop as they’re not clouded with overly generic content from every human on the planet.
wow...
Can anyone define what "cheat-code" means in terms of the save feature. Just looking for a little bit more detail to how well this works:)
It is admittedly untested experimentally but the idea is that if you are summarizing and stating the next step consistently it’s less likely to lose the thread of the conversation when it hits the context window. Anecdotally I’ve been able to extend a conversation well past the window and it reliably maintains its memory
You will have to update “Code Interpreter”, says Professor Don Gabo
how do i implement the /ts prompt
Do you have it defined in the commands section? If so just type and enter /ts during your chat
@@synapticlabsi.did try to do that it just kept telling me it didnt know it can you guide me a little more? Thank you so much
before trusting it to help me with my research, i challenged it to correctly present me with the prime factors of 7877.
surely, it can do such a simple thing; surely, i'm only misusing the prompt.
please make a video to explain how to get it to correctly answer something like that.
If you're trying to use LLMs for math, you're using a screw driver to hammer a nail. LLMs aren't good at math. Similarly research is not a good use of it unless you are feeding it the research you want to use as a basis, or maybe to help brainstorm some research questions, or support you in writing up the research based on your notes.
LLMs are a type of hyper advanced autocomplete based on their training data. You can get around this outside of chatgpt by connecting an API to a calculator, or a research database (you can use the arxiv plugin and scholarAI if you want), but just by itself it's going to get things wrong and hallucinate.
this prompt does not work correctly in Portuguese.
I tried to use it with commands in Portuguese, he understands, changes the language, but does not start the intelligent agents to help. Do you know how I can do to solve this?
ChatGPT isn't very good at any other language than english...I imagine you already translated the entire prompt into portuguese? You might want to add a rule that says "only respond in portuguese" or something...in portuguese
@@synapticlabs i translated, he responds in portuguese, the problem is that it doesn't perform the specific functions, it doesn't call another customized assistant for each task. I tried it on chatgpt 4 and it worked, however, I wanted to try it on 3.5, the turbo is faster
@@leonamnagel9999 yeah 3.5 is much faster but super unreliable happy to hear it worked!
Alot of these comments seem like bots but I’ll try it out
I hope not 🤖
I cant understand how people are still so low level at prompting. Like that prompt idea is many months old, we are ages ahead of that. How people don't know the real stuff?
My sister only discovered chat gpt yesterday so 🤷♀️
The clarifying questions are annoying because you spend all this time clarifying and then response is some generic non-sense that wastes the last 5 minutes of your interaction. Why does almost every interaction require a clarification. Its not clear what the clarification even does in terms of changing the answer if anything. It just seems like more interactive prompts that don't necessarily produce better results.
Sorry it’s not super useful to you but highly recommend you make some of your own rules to make it respond closer to the way you like. Tell it to ask fewer questions as well.