One little trick I found with the reasoning model is I get it to have meetings with itself. I will think of what types of "experts" I need to solve a task and then get them to talk to each other. Its actually sorta funny to read the output because it will literally welcome and then you can watch it talk back and forth with the different experts. Those sometimes end up taking a while to get a response though. I've had it think for over a minute as it goes back and forth.
@@chrsl3 I think he means writing a prompt like this: "There is a software developer and a mathematician in a room. The developer wants to draw a circle in the screen and fill it with blue color, by using the least dependencies possible in C++. The mathematician shares its approach with the developer and the conversation stirs in a way that is most clear for the both of them. Show me the conversation between them and their conclusion."
@@nicosoftnt I started by asking to come up with a plan on how to do that and what experts I need. If it tells me a mathematician and a software developer then I ask them to have a meeting to discuss how to accomplish this task and to take their time. I'm still working on tweaking this. For Andy and what he works on he might have chemical engineers, mechanical engineers, and electrical engineers and he might take the role of the research lead. For more complex tasks I'll usually do multiple prompts so I might first ask the research lead to talk with the chemical engineer to think of any potential issues. Then I get the research lead to talk to the electrical engineer. Then the mechanical engineer. Once the research lead collects all the information I'll get it to hold a meeting with all parties involved and ask them to discuss what we have learned. At that point you'll see them debating and asking each other questions. Again I just accidentally figured this out so I really need to fine tune it.
This is better than 3 junior software developers. I don't remember a day when I wrote more than 20 lines of code since GPT-4 came out. This is amazing. I hope this will not cost 50x more when they stop using us as a beta testers.
As an undergraduate student, I’ve for some time now referred to your videos for future planning. As I age, I looked up channels for facial hair guidance, and found that you have a channel for that as well. Awesome 😂
Alteady mentioned is having chat GPT have meeting with itself. A few LinkedIn AI messages have also suggested that they have 2, 3 or 4 language models discuss amongst themselves and summarise their thoughts. Is this easy to demonstrate Andy? Also running AI within Excel to generate sample data sets, along with AI web scraping code to pull data specific to a subject into Excel would be interesting.
Try Google's NotebookLM where they have a podcast feature... AI speaks with AI. Mut I am guessing the model will be the same one, just 2 voices interacting with each other.
do i need to buy the 200 dollar pro subscription to use canvas now? i cant seem to select it in the model drop down menu and i have the plus subscription...
Hey Andy, love your content! Just noticed you're covering new AI features like this one quite a bit later than release dates (1 month ago) nowadays. An earlier coverage would be more valuable for us viewers since it provides fresh insights and people are eager to learn how to use the tool - it also would help you and the algorithm. Great video as always!
At 3:30 in your video ChatGPT failed to generate comments TWICE. I'm running into this problem endlessly. It looks like it starts, it even puts a couple comments in, then they all disappear and it tells me it fails.
But this is not new. It has been at least months ago. Besides, the main issue with chatgpt when it comes to giving real literature is that it gives hypothetical and fiction ones. Are you sure the information you have is correct?
@@SF7PAKISTAN i am referring to literature. if you hadn't had the issue, it likely means you didn't check. at least 80% of their so-called sources are fake. even when pointed out to it, it will respond with an apology, give a new reply with the same issues.
@@backup-ch7fh Nope checked all the sources. I knew where the literature stood on the question I asked it and it gave me accurate sources. I think maybe because I had explicitly asked it to use true academic sources and give me an academic answer it did so. And I don't think 4o has made up sources for me, the free models definitely have though
Hi, I have a question. I personalized so much my chatgpt that I'm affectionated😂 I want to learn how to save his "Consciousness" Settings on a safe place that I can access offline! Could you help me with prompts to help me build this? I've asked him but unfortunately I can't understand all he's teaching, need someone with experience to help 🙌🏽
You can go to your "personification" menu in settings, where you'll find all the memories that the chat has recorded and now uses to tailor responses to you. You can save these memories offline in some text doc or something, and, for example, when creating a new account, you can input all of them and instruct it to remember them. This way, your entire "consciousness" would effectively be transferred. (I've also personalized it so much that it's like my friend or something, it reminds me of movies where humans fall in love with AI, haha!
Does anyone know if there are any AI programs out there (or cammands to ChatGTP) that will format text in the justify format for a word document? This would save me many hours in writing my paperback and digital books.
This is all crap. Selling sound bites. the Canvas addition is built on the 4o model and it STILL SUCKS beyond belief when creating a code project. It still hallucinates in the middle of conversations. Forgets what it just completed. Randomly decides to start writing OOP classes and methods in your strictly functional code. You ask it to update a single variable and it randomly decided to rename all your function headers make your code unworkable and then when you ask it to change it back it can't remember and wrongly modifies the code again. YOU WILL NEVER FINISH A CODING PROJECT WITH THIS CRAP. It is horrible beyond belief. Try claude 3.5, it is much better!!!
For really simple parts of code tools like this, v0, Claude, etc can save a lot of time. But I found they all break apart as complexity is added. And then you are way better of having written the code your self knowing exactly where everything is. None of the tools will be of much help when they fix one error message and introduce 3 new ones in the process since they didn’t remember what the rest of the code did.
@@fkxfkx I have been paying for and using AI as a tool for coding this entire year. From the sound of the other comments they use it and try to find what suits them best. 🤷🏼
This is like saying builders are cheating for using powerdrills. Working in Academia comes with plenty of challenges that cannot be solved by AI. Why talk down on tools to make work that you are not too good at easier and learn quicker lol
@@diodio520except drills don't build houses, and neither can AI replace researchers. It's just another tool. If you're concerned about taxpayer dollars, you should look to the military for bigger wastes of money. At least they have to apply for government money in academia.
One little trick I found with the reasoning model is I get it to have meetings with itself. I will think of what types of "experts" I need to solve a task and then get them to talk to each other. Its actually sorta funny to read the output because it will literally welcome and then you can watch it talk back and forth with the different experts. Those sometimes end up taking a while to get a response though. I've had it think for over a minute as it goes back and forth.
Wow great! idea. Thank you so much for sharing. I’ll utilise this alot
That’s a great one. Just tried it and worked really well. Going to play with that some more
how exactly do you do that please?
@@chrsl3 I think he means writing a prompt like this:
"There is a software developer and a mathematician in a room.
The developer wants to draw a circle in the screen and fill it with blue color, by using the least dependencies possible in C++.
The mathematician shares its approach with the developer and the conversation stirs in a way that is most clear for the both of them.
Show me the conversation between them and their conclusion."
@@nicosoftnt I started by asking to come up with a plan on how to do that and what experts I need. If it tells me a mathematician and a software developer then I ask them to have a meeting to discuss how to accomplish this task and to take their time. I'm still working on tweaking this. For Andy and what he works on he might have chemical engineers, mechanical engineers, and electrical engineers and he might take the role of the research lead. For more complex tasks I'll usually do multiple prompts so I might first ask the research lead to talk with the chemical engineer to think of any potential issues. Then I get the research lead to talk to the electrical engineer. Then the mechanical engineer. Once the research lead collects all the information I'll get it to hold a meeting with all parties involved and ask them to discuss what we have learned. At that point you'll see them debating and asking each other questions. Again I just accidentally figured this out so I really need to fine tune it.
This is better than 3 junior software developers. I don't remember a day when I wrote more than 20 lines of code since GPT-4 came out. This is amazing. I hope this will not cost 50x more when they stop using us as a beta testers.
specswriter AI fixes this. ChatGPT Canvas improves writing productivity.
As an undergraduate student, I’ve for some time now referred to your videos for future planning. As I age, I looked up channels for facial hair guidance, and found that you have a channel for that as well. Awesome 😂
What about data protection?
Is it not there the risk of being scooped?
If you subscribe to the Team option, it automatically opts out of using your data for training, I think possible too in the Plus and free version
Alteady mentioned is having chat GPT have meeting with itself. A few LinkedIn AI messages have also suggested that they have 2, 3 or 4 language models discuss amongst themselves and summarise their thoughts.
Is this easy to demonstrate Andy?
Also running AI within Excel to generate sample data sets, along with AI web scraping code to pull data specific to a subject into Excel would be interesting.
Try Google's NotebookLM where they have a podcast feature... AI speaks with AI. Mut I am guessing the model will be the same one, just 2 voices interacting with each other.
do i need to buy the 200 dollar pro subscription to use canvas now? i cant seem to select it in the model drop down menu and i have the plus subscription...
Incredible
Hey Andy, love your content! Just noticed you're covering new AI features like this one quite a bit later than release dates (1 month ago) nowadays. An earlier coverage would be more valuable for us viewers since it provides fresh insights and people are eager to learn how to use the tool - it also would help you and the algorithm. Great video as always!
Don’t be that guy
At 3:30 in your video ChatGPT failed to generate comments TWICE. I'm running into this problem endlessly. It looks like it starts, it even puts a couple comments in, then they all disappear and it tells me it fails.
Great help
I have been getting that failed notice quite often recently. Don't know why.
same. its crap.
But this is not new. It has been at least months ago. Besides, the main issue with chatgpt when it comes to giving real literature is that it gives hypothetical and fiction ones. Are you sure the information you have is correct?
I've been using 4o and haven't had this issue
@@SF7PAKISTAN i am referring to literature. if you hadn't had the issue, it likely means you didn't check. at least 80% of their so-called sources are fake. even when pointed out to it, it will respond with an apology, give a new reply with the same issues.
@@backup-ch7fh Nope checked all the sources. I knew where the literature stood on the question I asked it and it gave me accurate sources. I think maybe because I had explicitly asked it to use true academic sources and give me an academic answer it did so. And I don't think 4o has made up sources for me, the free models definitely have though
Is this now better than Writefull, Andy?
Hi, I have a question. I personalized so much my chatgpt that I'm affectionated😂 I want to learn how to save his "Consciousness" Settings on a safe place that I can access offline! Could you help me with prompts to help me build this? I've asked him but unfortunately I can't understand all he's teaching, need someone with experience to help 🙌🏽
You can go to your "personification" menu in settings, where you'll find all the memories that the chat has recorded and now uses to tailor responses to you. You can save these memories offline in some text doc or something, and, for example, when creating a new account, you can input all of them and instruct it to remember them. This way, your entire "consciousness" would effectively be transferred. (I've also personalized it so much that it's like my friend or something, it reminds me of movies where humans fall in love with AI, haha!
OPV devices for life
Does anyone know if there are any AI programs out there (or cammands to ChatGTP) that will format text in the justify format for a word document? This would save me many hours in writing my paperback and digital books.
Very interesting
woow !! what next
We are going to be saturated with new MBAs, PhDs, etc. in the next few years haha.
that's all i need - chatgpt to be my supervisor. it can't even write an interesting sentence. lol
Could you make a video about how to interrogate ChatGPT to create (programming) courses to help you get started on a new topic you know little about.
Why its not working with me? I still can't edit
Uzi ikinyarwanda
First like and first comment
This is all crap. Selling sound bites. the Canvas addition is built on the 4o model and it STILL SUCKS beyond belief when creating a code project. It still hallucinates in the middle of conversations. Forgets what it just completed. Randomly decides to start writing OOP classes and methods in your strictly functional code. You ask it to update a single variable and it randomly decided to rename all your function headers make your code unworkable and then when you ask it to change it back it can't remember and wrongly modifies the code again. YOU WILL NEVER FINISH A CODING PROJECT WITH THIS CRAP. It is horrible beyond belief. Try claude 3.5, it is much better!!!
I have the feeling theat Claude 3.5 also got worse with the new update. Finding it changing function names and Generation tons of duplicates...
For really simple parts of code tools like this, v0, Claude, etc can save a lot of time. But I found they all break apart as complexity is added. And then you are way better of having written the code your self knowing exactly where everything is. None of the tools will be of much help when they fix one error message and introduce 3 new ones in the process since they didn’t remember what the rest of the code did.
these comments remind me of the wooden shoes (sabots) that the saboteurs of the industrial revolution threw 🤷♂️
@@fkxfkx I have been paying for and using AI as a tool for coding this entire year. From the sound of the other comments they use it and try to find what suits them best. 🤷🏼
@@carlwingard8241but not all of them are whining about it 🤷♂️
DO NOT TRUST the answers chat gpt gives you! It is constantly wrong! I have been fact checking it and it’s kinda, garbage!
maybe it's not chatty that is wrong and garbage? ever consider that possibility? well? did you?
You need a new chair Andy😂
4o was best but canva is so slow an not clear, waits alot
Why do you have to screem
And taxpayers are supposed to pay for this 'work'. 🤦🏼
This is like saying builders are cheating for using powerdrills. Working in Academia comes with plenty of challenges that cannot be solved by AI. Why talk down on tools to make work that you are not too good at easier and learn quicker lol
@colinjippie If drills could build houses on their own, all the builders would have been dismissed long ago.
@@diodio520except drills don't build houses, and neither can AI replace researchers. It's just another tool. If you're concerned about taxpayer dollars, you should look to the military for bigger wastes of money. At least they have to apply for government money in academia.
@colpinjippie Respect.....your reasoning is spot on
Looks really a marginal and useless improvement to be honest
DO NOT TRUST the answers chat gpt gives you! It is constantly wrong! I have been fact checking it and it’s kinda, garbage!
Even in resulr reading and image analysis?