I like Canvas a lot. It's better than Claude's artifacts in that you can edit it, right there. However, Claude still writes better working code, from my experience. Clause just makes you work harder because it gives you code changes and you have to find out where to cut+paste to put it back together...
I only use Claude for coding. especially complex-ish codes. gpt4o, O1 they all failed time and time again so i stopped even bothering to use them for coding. with Claude projects + Sonnet 3.5 you could code anything. and its becoming smarter and better at coding evrymonth or 2. by next year new year im Very confident coding agents will be Trully useful and much better than now( claude will still be king, those agents will be run by claude llm’s)
Hey Matthew. I love my custom GPTs. In my opinion, it's a bit of pdf context and prompt engineering on steroids. I am taking online classes to finish my degree, and most of the instructors will drop all the material, tell you when everything is due and that it! No interaction or "actual teaching." My Custom GPTs become my expert personal Tutors (I don't share them). On the drive to work I can have a conversation with them and have them give me a 5 to 10min lecture. It's awesome~!
Similarly, I drop in PDFs, operators manuals, or procedures for certain equipment and ask my GPT to walk me through using it. My custom GPT has my company, personal job details and requirements, internal processes and such stored. I have to read new repair procedures or theory of operation documents all the time, so this is when I start a new chat. The GPT explains it to me the way I need to hear it for this specific job/chat.
We have built a web application generator CUSTOM GPT that generates the needed code to create complete solutions! To see it in action, you can google this: "car-ai-demo-short"
Running Python code in chat gpt is exceptional. I am going mad by watching one by one update from Open AI. Thank you so much Mathew I am watching your videos regularly, they are teaching me which I never knew before.
I don’t see any real threat to VS Code or other editors here. I find it hard to believe that OpenAI would even attempt to compete with real code editors in terms of overall functionality, as the developer experience they offer is nowhere near comparable. A code editor is so much more than just auto-complete.
Agreed, for now, but just imagine the power of each application and environment being agentically connected? With GPTPro reasoning and beyond, you can set the scope and leave the machine to come back with the results by autonomously iterating until it reaches the defined goal. Maybe it's at this point that we stop seeing code as the scope is all that's needed. Is this more than a year away do you think, or much further than that?
What you are failing to realize is that; This is opening the door for people who don't know how to code at all... Couple this with a crash course and all of a sudden...anyone will be able to code ..
@@DAFascend and it’s already happening. I used to code to augment publishing systems I managed 20 years ago (C++ and Perl) and I have not got the time/brains to learn Python well now, but, I know how to get around creating an environment, command line, file structures, etc that you still need. I’m already creating working Custom GPTs, run open source LLMs locally, created a fully operational agentic tool and I’ve barely whiten a line of code. In 12-15 months, I won’t need to do anything to spin this and far more powerful stuff up. Maybe not enterprise level stuff, but it’s all up for grabs.
@@DAFascend You’re missing my point. I’m not saying OpenAI won’t improve their models for coding or make it easier for people who don’t know how to code, they definitely will. What I’m saying is that I don’t see them competing directly with code editors or building a complete IDE. The market for that is too specific, too small, and too expensive to justify the effort.
@@mykhaylotymofyeyev9565 Fair point... I think you might be missing the point of AGI... They want to get to a point where AI will solve problems like these....with emergence. Trying to create truly intelligent machines... And my bet is they'll succeed before 2029 if not then sooner.
ChatGPT has becoming worse for each new release. They are trying to compensate with “new features”. I’m now using Claude in 90% of cases and only use OpenAI just to see how they would do it. Not surprised as key people left and only noobs and sales stayed.
Hi Matt, Custom GPTs helped us BIG TIME! We have a custom GPT that knows how to create the low-code for our data-driven Enterprise Software Development App. This means that by typing a prompt, functional parts for the ERP Solution are added at run-time! This opens the path to even faster solutions delivery!
11:24 This is like having a virtual Professor close to you to guide you through a lesson. Getting full comprehensive tips and comments on the overall story or lesson is excellent 👌🏾👌🏾👌🏾. 15:47 Canvas is making Chatgpt very interactive by providing multiple functionality into a single open window. This design setup provides efficiency with multiple powerful tools in one location (viewable in one place). 18:00 I can't believe that they've already incorporated multiple agents in OpenAi to assist each other in solving problems. 18:00 This is like having multiple professors in a round table brainstorming and fixing infinite problems. This is awesome 😎👍🏾.
It sounds like it only works if you have a job or poject where a lot of the work involves either exploring frontier AI capabilities or leveraging very complex AI functionalities. I'd only consider it for exploring the frontiers of AI coding without the current quota constraints but it's a hard bet that may flop a few months into putting it to the test
These features are temporarily available on the free plan. You know, like the crack dealer on the street corner... "Here try this, it's free. If you don't like it, you don't owe me anything. If you do, you know where to get more..."
I haven’t really used custom GPTs, but I use projects in Claude a lot. It’s handy to give it custom instructions and data to really keep the focus on whatever your goal is. I make new ones for almost any bigger idea I have.
im paying 200$ a month for the "pro" .. and i don't have access to SORA. But i watched some of the videos.. the UI/UX is impressive. What i dont believe at all is that they were able to make this when they showed it... like, IMHO at that day.. they just had a CGI, and Meta with kamaleon, sam, etc.. turned that CGI into what we see now "as sora". "OpenSource" as source of zilions dollars enterprises with 0 obligation to even cite the authors.. the next level of corporative piracy.
I utilize a created 'Identity' with most AI. Simply defined to enhance logic and reason. I haven't had to worry about 'hallucinations' in years. Works with all LLM not just ChatGPT.
My custom GPT that I have made and remade several times helps me run and program my CNC machine tool. The back story is, I purchased a high end very complicated CNC tool from Mazak. it came with a large stack of PDF user manuals covering how to use the machine, how to program, and how to repair and troubleshoot. total pages is something around 10,000. The challenge is my PDFs are very filled with diagrams and charts and pictures of parts of the machine. Last year the GPTs could not even read or look at the pictures. Now it looks like 4o can now do that. I am having a difficult time finding out enough information on how to create my custom GPT optimally. But I'm going to continue plugging away.
Great video Matt! I create an anime panel generator, its called MangaVerse and after it organizes the plot of the storyboard it will create a panel with multiple images based on the storyboard outline.
Custom GPT Example. I have a few companies. One of them has to follow specific state guidelines. I uploaded text documents that contain all the rules and regulations that the state requires. If I ever have a question on how to do somehting, chat can give me a guideline and a plan that meets all the state requirements.
until it forgets and then you are in hot water. if there is one thing my too many thousands of hours messing with these things has taught me is that the thing AI does best is forget. Don't get complacent and assume its telling you the truth. always double check.
Custom GPT example, I have a personal lawyer trained on legal docs I’ve signed, context on how I’ve approached different legal negotiations, legal form templates I’ve used, etc
ChatGPT 4.0 has been having internal Python console for almost a year! I remember, I had only my phone with me and I gave it some homework - to write me 10 functions, each of which had to show a particular geometric shape, with ASCII symbols, on the console. Then I asked it to show me the results (meaning the code for each sub) but instead it showed me a printout of the result of each run! That's how I knew it has an internal console and it can execute code on its own.
Our custom GPT understands our Enterprise App with built-in IDE. The GPT can create the low-code because it knows all the over 400+ stored procedures. The Custom GPT effectively builds the complete solution withT-SQL only, it even builds the UI with Modal Interactive pages, CRUD cards, Searching and Action Buttons.
Can you do a video on Gemini 2.0 flash and it's abilities because they are awesome and there are some interesting things also brooooo! Like native image generation and audio Generation and screen share and live audio and video streaming broooo!
I created my first ChatGPT at the weekend to help me develop my game. It has a lot of stories and every time I generate missions, I added it to its library as well as helping it provide me prompts to generate characters using flux
I started getting inline comment suggestions a few days back, that I had not asked for and they were really useful as well. This kind of stuff is a great example of AI products and feature integration that's going to take the world by storm in the next 12-18 months. All the fragmented apps and solutions we currently wrestle with connected to more capable agentic programs and tools are going to make average Joes (like me) look like tech wizards...
I just use CustomGPTs tailored to my work's knowledge base and loaded with the templates I use for m work outputs. I also use it for creative writing, health advice, and other things.
So basically you just replay the openAI videos but the only added value is that we get tons of sponsors , ads and commercials? Come on man. You used to be better than that. I'll give it a last chance. If the next video is like that I'll unsubscribe from your RUclips channel and discord. Thanks.
You should really ask for your money back oh wait you don’t pay anything for this content! I find it amazing that little socialist Marxist complain about people contributing content on RUclips when they don’t do anything themselves or pay for anything. . quit complaining he provides tons of content
dude what's wrong with you not everyone is subscribed to openai and has time to go watch their videos his channel covers Ai news if you don't wanna watch the sponsors skip it simple
That's not the point of his video. It's that he finds all the interesting things for you, and gives his commentary on it as a knowledgeable person with experience in this business. I think you appreciate him not enough
This tone of NEW (but not acknowledging not new to frontier models at large) is uncomfortable. The google docs, that everyone has been tweeting at them forever, I think is rad. BUT I’m also starting to think that Anthropic is playing the old school Apple role, where they sit back, watch what happens, and then release when they perfect things. Like I sort of don’t understand o1 when chained thought has been a thing for Claude since before MCP even - it has even been in the normal chats for a month now and they just didn’t make a big hype about it because, hell, even Perplexity Pro has been doing it for multiple months now, just not fully autonomously. I guess I’m just a little bias right now since Claude fully took over my Notion.
Hi Matt, we created a CUSTOM GPT that allows us to create complete enterprise business solutions. We made this possible by building a t-sql based data driven application framework.
GPTs and it’s equivalent on Claude (projects) are useful for giving them a list of files with examples and extra instructions or even files to actually develop. I have used both and sometimes think that I might end up having tens if not hundreds each for a specific part of my programming projects. I only wish they had more automation interfaces to them such ftp or API at least for the paid version. I’m subscribed to both.
This is very cool, but I'm really missing a branching chat system with the ability to link parts of conversations. The current OpenAI system, just a simple list of chats without even folder organization, feels insufficient for managing context effectively.
1. Loses code when editing. 2. Often cuts off the ends of the program. 3. After a web search it does not correct in canvas but in chat. 4. Code execution? How to install tensorflow?
I have used Custom GPTs in several areas. 1. I teach at MSU. I use a Custom GPT to grade written assignments. I upload the rubric, provide an ideal example assignment and provide instructions on what type of feedback to provide the students including the number of points. It is tougher on points than I would be but it does provide a lot of good feedback to the students that I could not. 2. I created a Custom GPT, Ask Doc, with all the content from my lectures and other works that incorporate my ideas and philosophy on leadership. The idea is students or clients could Ask Doc a question they are have about leading their team and the custom GPT would give the type of advice I would give. I think it is working well. I plan on sharing it with others soon to see what they think. 3. I have not started this yet, but I want to create a similar type Custom GPT, Ask Grandpa, for my grandkids that would share with them my life experiences and knowledge. Lessons I have learned. Experiences I have had, etc. A lot of the content in this will be stories from my life. Let me know if you’d like to discuss these 1-on-1.
Interesting examples... Re: Ask Grandpa Using voice-to-text dictation would allow you to share a lot of detail across many life experiences, randomly, as quick as you can speak about them. Then AI can sort, categorise, and merge common thoughts to complete those experiences. Like creating a mental time capsule... very clever!
We created a custom gpt (and also a Claude project) that creates the low code needed to create complete web apps with security, RBAC, CRUD, reporting, mailing and Modal pages to interact with users; all from the ide/gpt
I don't really use custom GPTs but I want to learn more about how to create them. I would like to be able to feed a few calculus books to "train" the custom GPT on, and then have a smarter LLM for calculus problems.
@@brianWreaves Thank you. Yes, you are 100% correct. But nonetheless I would "I want to learn more about how to create them..." Calculus just happens to be my current use case because I have the pdfs to use for RAG. I'm suggesting a video for our host. I'm just a sort of big dummy when it comes to these fancy pants LLMs and I like Matthew's videos since he makes me feel less dumb. :)
Having had a play their implementation of canvas lacks many basic capabilities, which you don’t mention. You cannot save the Canvas to it’s own thread or project as you can with Anthropic. I asked it to create a google compatible doc of a canvas - it provided a word doc but half of it was missing when I opened it. On the second attempt it did however create a markdown which I could then copy and paste into my obsidian notebook. However, i would like the option to save a copy of the canvas to a folder in chatgbt rather than it being embedded in a chat thread all the time
After chatting about Canvas from within Canvas, it sounds like even though WASM is being used, it is not running locally but instead on OpenAI servers and interactive code is not allowed (at least graphics apps, like PySide6 or sliders in Matplotlib). Still, the PySide6 version of my 6th degree Bezier editor did work after one iteration of bug fixes and running the code in my local IDE (VS Code). I tried and failed to Matplotlib to work, but it only renders an image, so no interactivity. I then had it create a TypeScript + React version which I have yet to try (once again it would have to be in my local IDE). Python is the only supported language for running within Canvas. Even so, seeing the Canvas code being edited line-by-line was pretty neat even though it always rewrites the whole file line by line in place.
So all they did with canvas was take it out of preview mode and integrated it into base 4o. Nothing was fixed or changed - can't rename or delete a canvas and truncating code is still a major issue for any real project.
Thanks for the video. I still feel like they are behind on the code editor part of it. I am thinking it is just a limitation of the chat paradigm their ui is built around vs say lovable.dev and bolt.new that is, yes chat, but the tool is really about building websites and apps. Then with Cursor that you can work on the code based when things get tricky is just more a focused tool than what they might be able to do in the web like scripts, more complex backend code, playwright etc.
It would be nice if Canvass could help create a custom GUI, placing text boxes, input boxes, buttons, etc, with drag and drop onto the canvas. Then, have it identify and name each object and the power to ascribe attributes and functions to the objects on the GUI. Once it creates the GUI, you can move the objects around, add additional functionality, and run the code and test.
Canvas run code really didn’t work very well, at for that to be a huge thing is a joke 😂 And for 200$ + 25% taxes, and struggling to make it do a relatively simple task, is so not satisfying 🤨
It's not, the API is for that. This is for hobbyists, simple projects and best of all learning. I'm not a programmer by trade, but I've been getting into learning code the past few years as a hobby and best of all doing code as a hobby helps my mind and critical thinking. It's like exercise for your mind, I'm addicted to it and learning even if I don't go all the way into Enterprise. This tool is perfect for that. I'll probably be subscribing now.
Platform risks will always exist, but there is no way OpenAI can continue to both develop foundational models and provide the product. No matter how smart they are, they are only so many of them and there is a lot more smart peoplpe outside. As a developper developping on top of AI models and being at platform risk, you have to know which one is the flashlight app (platform risk) and which one is the next billion dollar company (like Facebook or Uber).
I do not agree. By personally creating a custom gpt that created custom low-code for our framework, now the gpt produces working functionality for our t-sql app much faster that any programmer could ever do!
Custom GPTs can be great. Just be prepared to train it thoughtfully. Our custom gpt talks the language of our application framework and now the gpt spits out complete apps
They do this on purpose. It should make you wonder WHY they CHOOSE to do it like that, what impression they want to convey and why they want to convey it.
@@technolus5742 Yep: Tech companies might present their newest features in a seemingly "silly" podcast format for several strategic reasons: 1. Audience Engagement Podcasts are an increasingly popular medium with a diverse and engaged audience. A casual, entertaining approach can make technical or dry content more appealing to a broader audience. 2. Humanizing the Brand Humor and informal formats make the company seem more relatable and approachable. This helps break down the perception of tech companies as impersonal corporations. 3. Targeting a Younger Demographic Younger audiences often value content that feels authentic, fun, and less corporate. A playful podcast might resonate better with these groups. 4. Differentiation in a Crowded Market With so many companies launching new features, a unique and unconventional format helps stand out and makes the announcement memorable. 5. Viral Potential A humorous or quirky presentation is more likely to be shared on social media, giving the feature and the company additional exposure. 6. Showcasing Creativity Presenting in a novel way highlights the company's creativity and innovation, aligning with their branding as forward-thinking. 7. Storytelling Opportunity Podcasts provide a format where companies can not only announce a feature but also explain the story behind it, its purpose, and its benefits in a conversational tone. 8. Cost-Effectiveness Producing a podcast is relatively inexpensive compared to more traditional marketing campaigns, yet it can still reach a wide audience effectively.
I find them too painful to watch. They repeat each others pointless jargon. The videos OpenAI released showing how to use SORA were short, to the point and felt impactful.
interesting, but not useful yet. The entire copy paste prompt copy paste run workflow is a dog for real software projects. The reason Cursor/Windsurf are appealing is that you can @ mention any portion of your code into the context, the changes are logged in your git, you can walk through each change, and use a real debugger to see the application state. The debugging process using a browser based code tool like canvas usually devolves into inserting logging statements throughout the code to hopefully catch the state of that problematic thing.
they are spending absurd amounts of time and effort on this junk chat interface they have that is in all aspects subpar to the free and opensource alternatives. the ONLY thing that makes this not dead in the water is that its natively attached to a good LLM model. if they have shifted from all of their resources working on improving the AI to spending a ton of money and time trying to play catch up on the interface then that should speak volumes about whats going on inside openAI. we can already plug any LLM we want into a much better interface using API...
You still won’t get in, but instead of saying not available in your region, it will have the ‘access suspended’ message instead, as it’s overloaded for new users.
I like Canvas a lot. It's better than Claude's artifacts in that you can edit it, right there. However, Claude still writes better working code, from my experience. Clause just makes you work harder because it gives you code changes and you have to find out where to cut+paste to put it back together...
Lol. I max out Claude all the time, then only use GPT when I am waiting for it to recharge.
Claude is still ahead in this like you said. I do hit limits though which is annoying. I still prefer it over ChatGPT as of right now.
I only use Claude for coding. especially complex-ish codes. gpt4o, O1 they all failed time and time again so i stopped even bothering to use them for coding.
with Claude projects + Sonnet 3.5
you could code anything. and its becoming smarter and better at coding evrymonth or 2. by next year new year im Very confident coding agents will be Trully useful and much better than now( claude
will still be king, those agents will be run by claude llm’s)
Awesome vid, I like codebuff! Are you planning on making a video about Hoody AI for us poor students?
Mistral AI Le Chat has been offering a similar interactive canvas for a few weeks now. And it's free.
So are artifacts on censored Claude
Hey Matthew. I love my custom GPTs. In my opinion, it's a bit of pdf context and prompt engineering on steroids. I am taking online classes to finish my degree, and most of the instructors will drop all the material, tell you when everything is due and that it! No interaction or "actual teaching." My Custom GPTs become my expert personal Tutors (I don't share them). On the drive to work I can have a conversation with them and have them give me a 5 to 10min lecture. It's awesome~!
Similarly, I drop in PDFs, operators manuals, or procedures for certain equipment and ask my GPT to walk me through using it. My custom GPT has my company, personal job details and requirements, internal processes and such stored. I have to read new repair procedures or theory of operation documents all the time, so this is when I start a new chat. The GPT explains it to me the way I need to hear it for this specific job/chat.
I made one for tax advice. Unfortunately it limited me to only a small subset of the IRS publications. Maybe this has improved since then.
We have built a web application generator CUSTOM GPT that generates the needed code to create complete solutions! To see it in action, you can google this: "car-ai-demo-short"
Running Python code in chat gpt is exceptional. I am going mad by watching one by one update from Open AI. Thank you so much Mathew I am watching your videos regularly, they are teaching me which I never knew before.
I don’t see any real threat to VS Code or other editors here. I find it hard to believe that OpenAI would even attempt to compete with real code editors in terms of overall functionality, as the developer experience they offer is nowhere near comparable. A code editor is so much more than just auto-complete.
Agreed, for now, but just imagine the power of each application and environment being agentically connected? With GPTPro reasoning and beyond, you can set the scope and leave the machine to come back with the results by autonomously iterating until it reaches the defined goal. Maybe it's at this point that we stop seeing code as the scope is all that's needed. Is this more than a year away do you think, or much further than that?
What you are failing to realize is that; This is opening the door for people who don't know how to code at all... Couple this with a crash course and all of a sudden...anyone will be able to code ..
@@DAFascend and it’s already happening. I used to code to augment publishing systems I managed 20 years ago (C++ and Perl) and I have not got the time/brains to learn Python well now, but, I know how to get around creating an environment, command line, file structures, etc that you still need. I’m already creating working Custom GPTs, run open source LLMs locally, created a fully operational agentic tool and I’ve barely whiten a line of code. In 12-15 months, I won’t need to do anything to spin this and far more powerful stuff up. Maybe not enterprise level stuff, but it’s all up for grabs.
@@DAFascend You’re missing my point. I’m not saying OpenAI won’t improve their models for coding or make it easier for people who don’t know how to code, they definitely will. What I’m saying is that I don’t see them competing directly with code editors or building a complete IDE. The market for that is too specific, too small, and too expensive to justify the effort.
@@mykhaylotymofyeyev9565 Fair point... I think you might be missing the point of AGI... They want to get to a point where AI will solve problems like these....with emergence. Trying to create truly intelligent machines... And my bet is they'll succeed before 2029 if not then sooner.
ChatGPT has becoming worse for each new release. They are trying to compensate with “new features”. I’m now using Claude in 90% of cases and only use OpenAI just to see how they would do it.
Not surprised as key people left and only noobs and sales stayed.
Hi Matt, Custom GPTs helped us BIG TIME! We have a custom GPT that knows how to create the low-code for our data-driven Enterprise Software Development App. This means that by typing a prompt, functional parts for the ERP Solution are added at run-time! This opens the path to even faster solutions delivery!
11:24 This is like having a virtual Professor close to you to guide you through a lesson. Getting full comprehensive tips and comments on the overall story or lesson is excellent 👌🏾👌🏾👌🏾. 15:47 Canvas is making Chatgpt very interactive by providing multiple functionality into a single open window. This design setup provides efficiency with multiple powerful tools in one location (viewable in one place). 18:00 I can't believe that they've already incorporated multiple agents in OpenAi to assist each other in solving problems. 18:00 This is like having multiple professors in a round table brainstorming and fixing infinite problems. This is awesome 😎👍🏾.
The fifth day should be big because in the song it’s five golden rings😂
Yeah no, that's just a reference to ring necked pheasants 🤫
Im just checking out till late next week lol
grammarly is gone!
They are probably already dying
Still scratching my head now as to why I need to stick the plus subscription
It sounds like it only works if you have a job or poject where a lot of the work involves either exploring frontier AI capabilities or leveraging very complex AI functionalities. I'd only consider it for exploring the frontiers of AI coding without the current quota constraints but it's a hard bet that may flop a few months into putting it to the test
Sora
literally zero reason. OpenAI is simply making the case for Open source. There is nothing they've released that isnt better opensourced
These features are temporarily available on the free plan. You know, like the crack dealer on the street corner... "Here try this, it's free. If you don't like it, you don't owe me anything. If you do, you know where to get more..."
@@Artorias920 i wish i could post this 10000 times on every stupid openai hype video i see. OPEN SOURCE EVERYTHING!
Canvas was a big bummer for me so I hope the updates are fixing all the issues
I haven’t really used custom GPTs, but I use projects in Claude a lot. It’s handy to give it custom instructions and data to really keep the focus on whatever your goal is. I make new ones for almost any bigger idea I have.
Claude projects is awesome but completely different than custom GPT’s.
Combining them would be more ideal than choosing 1
I tested a few random custom GPTs and they were basically useless.
@@wdmeister Some seem like they're nothing more than advertisements for products outside cGPT.
Love the new Canvas updates! Being able to run Python code directly in ChatGPT is a game changer for me. Can't wait to try it out!
It would not run the first code it created in canvas: "the Canvas environment here does not support executing Pygame or GUI-based scripts."
You’re right Matt, it seems the future is going to be a platform almost like adobe, creative cloud
im paying 200$ a month for the "pro" .. and i don't have access to SORA. But i watched some of the videos.. the UI/UX is impressive. What i dont believe at all is that they were able to make this when they showed it... like, IMHO at that day.. they just had a CGI, and Meta with kamaleon, sam, etc.. turned that CGI into what we see now "as sora". "OpenSource" as source of zilions dollars enterprises with 0 obligation to even cite the authors.. the next level of corporative piracy.
You can use vpn
I utilize a created 'Identity' with most AI. Simply defined to enhance logic and reason. I haven't had to worry about 'hallucinations' in years. Works with all LLM not just ChatGPT.
Nice. Can you explain more please, and how to do it? I am sure others would like it also!
Thanks
My custom GPT that I have made and remade several times helps me run and program my CNC machine tool. The back story is, I purchased a high end very complicated CNC tool from Mazak. it came with a large stack of PDF user manuals covering how to use the machine, how to program, and how to repair and troubleshoot. total pages is something around 10,000. The challenge is my PDFs are very filled with diagrams and charts and pictures of parts of the machine. Last year the GPTs could not even read or look at the pictures. Now it looks like 4o can now do that. I am having a difficult time finding out enough information on how to create my custom GPT optimally. But I'm going to continue plugging away.
Great video Matt! I create an anime panel generator, its called MangaVerse and after it organizes the plot of the storyboard it will create a panel with multiple images based on the storyboard outline.
Custom GPT Example. I have a few companies. One of them has to follow specific state guidelines. I uploaded text documents that contain all the rules and regulations that the state requires. If I ever have a question on how to do somehting, chat can give me a guideline and a plan that meets all the state requirements.
until it forgets and then you are in hot water. if there is one thing my too many thousands of hours messing with these things has taught me is that the thing AI does best is forget. Don't get complacent and assume its telling you the truth. always double check.
Sounds like a great use case for NotebookLM. It only draws upon provided artefacts and doesn't hallucinate.
Custom GPT example, I have a personal lawyer trained on legal docs I’ve signed, context on how I’ve approached different legal negotiations, legal form templates I’ve used, etc
ChatGPT 4.0 has been having internal Python console for almost a year! I remember, I had only my phone with me and I gave it some homework - to write me 10 functions, each of which had to show a particular geometric shape, with ASCII symbols, on the console. Then I asked it to show me the results (meaning the code for each sub) but instead it showed me a printout of the result of each run! That's how I knew it has an internal console and it can execute code on its own.
Great - except 80%+ of the web runs on PHP - and not Python
custom GPTs are amazing they ve saved me thousands of euros. especially lawyer GPTs
Our custom GPT understands our Enterprise App with built-in IDE. The GPT can create the low-code because it knows all the over 400+ stored procedures. The Custom GPT effectively builds the complete solution withT-SQL only, it even builds the UI with Modal Interactive pages, CRUD cards, Searching and Action Buttons.
Platform risk is extremely high with all of this.
Yesterday ChatGPT Canvas, today ChatGPT Coma 😂
Can you do a video on Gemini 2.0 flash and it's abilities because they are awesome and there are some interesting things also brooooo! Like native image generation and audio Generation and screen share and live audio and video streaming broooo!
Now all we need is a good Open AI Information and Knowledge Cache for quick and easy AI augmented retrieval 😎🤖
I created my first ChatGPT at the weekend to help me develop my game. It has a lot of stories and every time I generate missions, I added it to its library as well as helping it provide me prompts to generate characters using flux
That’s so cool! Are you a game designer/engineer? I would love to learn how to do that
o1 GPTs with tools is gna be lit
200$ 🤑
their disadvantage here is that they relay on their own models, open source ui can easily dethrone this because of flexibility
I started getting inline comment suggestions a few days back, that I had not asked for and they were really useful as well. This kind of stuff is a great example of AI products and feature integration that's going to take the world by storm in the next 12-18 months. All the fragmented apps and solutions we currently wrestle with connected to more capable agentic programs and tools are going to make average Joes (like me) look like tech wizards...
Expect copilot to be implemented throughout VS code
The video is really cool. I'm watching someone watching a video and repeating what the people in the video are saying. Unbelievable
I just use CustomGPTs tailored to my work's knowledge base and loaded with the templates I use for m work outputs. I also use it for creative writing, health advice, and other things.
Custom GPTs are as cool as the tool calls they can leverage. 😌
That is correct that is exactly how we customized a gpt for our "t-sql dot app" framework. Now it spits out applications instead of code❤
@ got a link to that bad boy?
Diffs is what i've been missing; great to get that now.
So basically you just replay the openAI videos but the only added value is that we get tons of sponsors , ads and commercials? Come on man. You used to be better than that. I'll give it a last chance. If the next video is like that I'll unsubscribe from your RUclips channel and discord. Thanks.
thats like 20% of his videos, not all
still bad though
You should really ask for your money back oh wait you don’t pay anything for this content! I find it amazing that little socialist Marxist complain about people contributing content on RUclips when they don’t do anything themselves or pay for anything. . quit complaining he provides tons of content
the Sniperwolf of AI
dude what's wrong with you not everyone is subscribed to openai and has time to go watch their videos his channel covers Ai news if you don't wanna watch the sponsors skip it simple
That's not the point of his video. It's that he finds all the interesting things for you, and gives his commentary on it as a knowledgeable person with experience in this business. I think you appreciate him not enough
This tone of NEW (but not acknowledging not new to frontier models at large) is uncomfortable. The google docs, that everyone has been tweeting at them forever, I think is rad. BUT I’m also starting to think that Anthropic is playing the old school Apple role, where they sit back, watch what happens, and then release when they perfect things. Like I sort of don’t understand o1 when chained thought has been a thing for Claude since before MCP even - it has even been in the normal chats for a month now and they just didn’t make a big hype about it because, hell, even Perplexity Pro has been doing it for multiple months now, just not fully autonomously. I guess I’m just a little bias right now since Claude fully took over my Notion.
Windsurf acts like this, except it is local
Hi Matt, we created a CUSTOM GPT that allows us to create complete enterprise business solutions. We made this possible by building a t-sql based data driven application framework.
We are UX1 from NL
That’s awesome. Would love to learn more about that… what type of enterprise business solutions and what data do you feed the model?
@@jjtm7 you could google literally for "car-ai-demo-short" to see this in action
I've been using canvas since release and it is AMAZING.
They're still pushing Santa on the kids, huh?
GPTs and it’s equivalent on Claude (projects) are useful for giving them a list of files with examples and extra instructions or even files to actually develop. I have used both and sometimes think that I might end up having tens if not hundreds each for a specific part of my programming projects. I only wish they had more automation interfaces to them such ftp or API at least for the paid version. I’m subscribed to both.
I love using codebuff. I like how minimalistic it is, cursor is too heavy.
Canvas seems to be gone, I hope it come back, I loved it.
Matthew, please get some sleep my guy. Don’t stay up that late😢
We just need a place to set environment variables for Canvas and Ciao bellaa
This could have been Copilot in Word, which would’ve been awesome..
Claude user must feel like Android users for years, when Apple releases stuff. :D
This is very cool, but I'm really missing a branching chat system with the ability to link parts of conversations.
The current OpenAI system, just a simple list of chats without even folder organization, feels insufficient for managing context effectively.
1. Loses code when editing. 2. Often cuts off the ends of the program. 3. After a web search it does not correct in canvas but in chat. 4. Code execution? How to install tensorflow?
how do i run tkinter apps?
I really dont care about sora much but damn openai, you should have known the demand would have been this high, i would like to try it.
I have used Custom GPTs in several areas.
1. I teach at MSU. I use a Custom GPT to grade written assignments. I upload the rubric, provide an ideal example assignment and provide instructions on what type of feedback to provide the students including the number of points. It is tougher on points than I would be but it does provide a lot of good feedback to the students that I could not.
2. I created a Custom GPT, Ask Doc, with all the content from my lectures and other works that incorporate my ideas and philosophy on leadership. The idea is students or clients could Ask Doc a question they are have about leading their team and the custom GPT would give the type of advice I would give. I think it is working well. I plan on sharing it with others soon to see what they think.
3. I have not started this yet, but I want to create a similar type Custom GPT, Ask Grandpa, for my grandkids that would share with them my life experiences and knowledge. Lessons I have learned. Experiences I have had, etc. A lot of the content in this will be stories from my life.
Let me know if you’d like to discuss these 1-on-1.
Interesting examples...
Re: Ask Grandpa
Using voice-to-text dictation would allow you to share a lot of detail across many life experiences, randomly, as quick as you can speak about them. Then AI can sort, categorise, and merge common thoughts to complete those experiences. Like creating a mental time capsule... very clever!
We created a custom gpt (and also a Claude project) that creates the low code needed to create complete web apps with security, RBAC, CRUD, reporting, mailing and Modal pages to interact with users; all from the ide/gpt
Best practices into loading data into the gpt? Text, pdf, etc? What about charts and graphs? PowerPoints, etc?
Looks like OpenAI is playing catch-up with Claude.
I don't really use custom GPTs but I want to learn more about how to create them. I would like to be able to feed a few calculus books to "train" the custom GPT on, and then have a smarter LLM for calculus problems.
Quite likely there are already several calculous GPT available. Just search for them and test to see if they meet your needs.
@@brianWreaves Thank you. Yes, you are 100% correct. But nonetheless I would "I want to learn more about how to create them..." Calculus just happens to be my current use case because I have the pdfs to use for RAG. I'm suggesting a video for our host. I'm just a sort of big dummy when it comes to these fancy pants LLMs and I like Matthew's videos since he makes me feel less dumb. :)
Let me know if you find out how. I am also interested in how to create them…
This is great, but clearly copied from Claude with editing added
Having had a play their implementation of canvas lacks many basic capabilities, which you don’t mention. You cannot save the Canvas to it’s own thread or project as you can with Anthropic. I asked it to create a google compatible doc of a canvas - it provided a word doc but half of it was missing when I opened it. On the second attempt it did however create a markdown which I could then copy and paste into my obsidian notebook. However, i would like the option to save a copy of the canvas to a folder in chatgbt rather than it being embedded in a chat thread all the time
Don't shave until the 12 days are done 🤪
After chatting about Canvas from within Canvas, it sounds like even though WASM is being used, it is not running locally but instead on OpenAI servers and interactive code is not allowed (at least graphics apps, like PySide6 or sliders in Matplotlib). Still, the PySide6 version of my 6th degree Bezier editor did work after one iteration of bug fixes and running the code in my local IDE (VS Code). I tried and failed to Matplotlib to work, but it only renders an image, so no interactivity. I then had it create a TypeScript + React version which I have yet to try (once again it would have to be in my local IDE). Python is the only supported language for running within Canvas. Even so, seeing the Canvas code being edited line-by-line was pretty neat even though it always rewrites the whole file line by line in place.
You look tired brother. Get some rest. o1 will still be there tomorrow. :0P
So all they did with canvas was take it out of preview mode and integrated it into base 4o. Nothing was fixed or changed - can't rename or delete a canvas and truncating code is still a major issue for any real project.
vscode is Microsoft. They own significant shares in Openai, right? I think they'll just integrate into vscode...
Thanks for the video.
I still feel like they are behind on the code editor part of it. I am thinking it is just a limitation of the chat paradigm their ui is built around vs say lovable.dev and bolt.new that is, yes chat, but the tool is really about building websites and apps. Then with Cursor that you can work on the code based when things get tricky is just more a focused tool than what they might be able to do in the web like scripts, more complex backend code, playwright etc.
I've been executing code on my own machine from Claude using MCP. It's not a first.
It would be nice if Canvass could help create a custom GUI, placing text boxes, input boxes, buttons, etc, with drag and drop onto the canvas. Then, have it identify and name each object and the power to ascribe attributes and functions to the objects on the GUI. Once it creates the GUI, you can move the objects around, add additional functionality, and run the code and test.
Canvas run code really didn’t work very well, at for that to be a huge thing is a joke 😂 And for 200$ + 25% taxes, and struggling to make it do a relatively simple task, is so not satisfying 🤨
if you pay for Sora, you arent paying attention...
Bro your eyes are SWOLL. What happened?!
Extensions put cursor (a fork of vscode) lightyears ahead still
I just used canvas, and it sucked, at least for today. I asked to stop the bs and to return to a more traditional approach.
Maybe dropping Twitter in favor of something else, or others platforms. Something like Bluesky ?
Hi pls provide some use cases for learn the open ai operations
Dude your videos are longer than the actual release videos. Gets annoying pretty quick.
Great news
Technically hasn’t ChatGPT been able to run python with analyze for a long time?
Why would an enterprise developer use this?
It's not, the API is for that. This is for hobbyists, simple projects and best of all learning. I'm not a programmer by trade, but I've been getting into learning code the past few years as a hobby and best of all doing code as a hobby helps my mind and critical thinking. It's like exercise for your mind, I'm addicted to it and learning even if I don't go all the way into Enterprise. This tool is perfect for that. I'll probably be subscribing now.
Its like the code feature is very basic - maybe aimed at school kids?
Platform risks will always exist, but there is no way OpenAI can continue to both develop foundational models and provide the product. No matter how smart they are, they are only so many of them and there is a lot more smart peoplpe outside. As a developper developping on top of AI models and being at platform risk, you have to know which one is the flashlight app (platform risk) and which one is the next billion dollar company (like Facebook or Uber).
Does that mean I can turn my GPT into an app in the app stores by using Python?
Without a requirements.txt running python in useless. Hopefully it's coming.
basically they got nothing new or innovative and try to make up for it with fancy useless non features everyone had before them.
I don't use custom GPTs. Wolfram seemed the most promising but still doesn't work well. GPTs suck at the moment
I do not agree. By personally creating a custom gpt that created custom low-code for our framework, now the gpt produces working functionality for our t-sql app much faster that any programmer could ever do!
Custom GPTs can be great. Just be prepared to train it thoughtfully. Our custom gpt talks the language of our application framework and now the gpt spits out complete apps
Make video on XAI Grok with 100k gpu working simultaneously and future tu expend to 200k then to million gpus.
not working for me code runs successfully but doesn't display anything in the console.
It to bad that the canvas is unavailable to work with o1 models
i love how amateurish their presentations are - considering the world-changing nature... imagine if apple did this how it would look!
They do this on purpose.
It should make you wonder WHY they CHOOSE to do it like that, what impression they want to convey and why they want to convey it.
@@technolus5742 Yep:
Tech companies might present their newest features in a seemingly "silly" podcast format for several strategic reasons:
1. Audience Engagement
Podcasts are an increasingly popular medium with a diverse and engaged audience. A casual, entertaining approach can make technical or dry content more appealing to a broader audience.
2. Humanizing the Brand
Humor and informal formats make the company seem more relatable and approachable. This helps break down the perception of tech companies as impersonal corporations.
3. Targeting a Younger Demographic
Younger audiences often value content that feels authentic, fun, and less corporate. A playful podcast might resonate better with these groups.
4. Differentiation in a Crowded Market
With so many companies launching new features, a unique and unconventional format helps stand out and makes the announcement memorable.
5. Viral Potential
A humorous or quirky presentation is more likely to be shared on social media, giving the feature and the company additional exposure.
6. Showcasing Creativity
Presenting in a novel way highlights the company's creativity and innovation, aligning with their branding as forward-thinking.
7. Storytelling Opportunity
Podcasts provide a format where companies can not only announce a feature but also explain the story behind it, its purpose, and its benefits in a conversational tone.
8. Cost-Effectiveness
Producing a podcast is relatively inexpensive compared to more traditional marketing campaigns, yet it can still reach a wide audience effectively.
I find them too painful to watch. They repeat each others pointless jargon. The videos OpenAI released showing how to use SORA were short, to the point and felt impactful.
goes right in line with their amateurish products and interface, par for the course.
@@Axio-Flex lol peak dunning Krueger right here.
Thank you ^_^
interesting, but not useful yet. The entire copy paste prompt copy paste run workflow is a dog for real software projects. The reason Cursor/Windsurf are appealing is that you can @ mention any portion of your code into the context, the changes are logged in your git, you can walk through each change, and use a real debugger to see the application state. The debugging process using a browser based code tool like canvas usually devolves into inserting logging statements throughout the code to hopefully catch the state of that problematic thing.
they are spending absurd amounts of time and effort on this junk chat interface they have that is in all aspects subpar to the free and opensource alternatives. the ONLY thing that makes this not dead in the water is that its natively attached to a good LLM model. if they have shifted from all of their resources working on improving the AI to spending a ton of money and time trying to play catch up on the interface then that should speak volumes about whats going on inside openAI. we can already plug any LLM we want into a much better interface using API...
I wish you stop saying that Sora is available to everyone.
I hope this canvas update is the setup for GPT-4.5 or Orion.
Its for gpt 4o only. Again, very underwhelming
Nothing new for while. I can't see anything impressive.
Sora for everybody, buy not Europe or UK....
Log out of all ChatGPT instances, Delete browser caches cookies etc, log into a VPN with US access point, log back in and voila
You still won’t get in, but instead of saying not available in your region, it will have the ‘access suspended’ message instead, as it’s overloaded for new users.
@@HuffTheFluff Why should I? I pay the fee every month like everyone else, but always the last country on earth to receive anything new.
thx matt b! 🙏📈🇺🇸
Wow. I have been using Claude for a long time. OpenAI is making it really hard not to switch
pfff.. python
Open AI sponsored videos you are better then that be natural.
Hiding this channel now, I'm fully sick of all the AI stuff.