GPTs vs Assistants API - which one is best for you?
Вставка
- Опубліковано 5 чер 2024
- OpenAI recently announced two powerful new features: Custom GPTs and the Assistants API - but it's not always obvious when you should use one or the other! In this video I do an in-depth comparison of GPTs and the Assistants API to help you see the similarities, differences, and applications of the two new technologies.
Many videos on UA-cam may mislead you to believe that you can deploy Custom GPTs on your website, but this is only possible with the Assistants API. I clear up this and many other GPT nuances when it comes to using these new features announced at the OpenAI DevDay. This in-depth artificial intelligence comparison of the Assistants API and GPTs is perfect for both beginners and seasoned developers who want to explore the new capabilities of OpenAI!
📚 Resources:
- Octonote: octonote.com (worldbuilding app shown)
- OpenAI API Documentation: platform.openai.com/docs/assi...
- Monsterbrew GPT: chat.openai.com/g/g-PnTvElBRO...
🔥 Key Highlights
This video covers the following areas of comparison:
🧙GPT vs Assistants API Developer Experience:
- GPTs vs Assistants API overview
- GPTs capabilities and developer experience
- Assistants API capabilities and dev experience
- Knowledge capabilities in GPTs and the API
- Knowledge refresh and uploading files using the Assistants API
- Assistants API limitations such as image generation and vision
- Tools & Actions inside GPTs
- Function calling with the Assistants API
- Conversation context via threads in the Assistants API
- Analytics for GPTs vs the Assistants API
💻GPT vs Assistants API User Experience:
- Your users needing a GPT Plus account for GPTs
- UI limitations for GPTs
- Possible UIs using Assistants API
🛠️GPT vs Assistants API Maintenance
- Simple maintenance for GPTs
- Hosting and deployment needs for Assistants API
- Assistants API costs
💡GPTs Usecases and Assistants API Usecases
We go through a number of example usecases that help you get a feel for when to use GPTs or the Assistants API
- Plant ID app
- Adding a GPT-powered chatbot to your site
- Spotify DJ
- Worldbuilding app using GPTs
If you are not sure which technology to use for your usecase, drop a comment down below and I'd love to help you out! 🔮
🚀 What You'll Learn:
- The difference between GPTs and Assistant API
- When to use one or the other
- Limitations of GPTs
- Limitations of Assistants API
- Capabilities of GPTs
- Capabilities of Assistants API
- That you can't use custom GPTs on your website
- That your GPT users need GPT Plus
💡 Perfect for Viewers Interested in:
- Artificial Intelligence
- Understanding differences between OpenAI GPTs and the API
- AI-powered chatbots
- Building custom GPTs
- Building apps using the Assistants API
- Using Actions inside of custom GPTs
- Adding knowledge to your custom GPTs
- Chatting with your files in GPT
- OpenAI API applications
- Building custom GPTs
- Making money with GPTs
- New OpenAI capabilities
- Learning about the latest AI tech
Subscribe for more comprehensive deep-dives and tutorials on AI and programming! Stay up to date on the latest AI updates and news!!
💬 Questions or Feedback? Drop your thoughts in the comments below, and I'll be sure to get back to you!
#openai #devday #ChatGPT #ai #aicommunity #gpt #Gpt4 #OpenAIStore #DallE3 #ChatGPT4Turbo #OpenAIAssistants #OpenAIAgents #CustomGPT #OpenAIAPI #OpenAIDevDay #AIChatbot #OpenAITutorial #AIIntegration #ChatbotTutorial #TechTutorial #CustomGPTs #NewAppStore #OpenAIStore #GPTStore #ChatGPTStore #CustomGPTTutorial #GPTTutorial #CustomGPTs #OpenAIMarketplace #GPTMarketplace #GPTActions #GPTKnowledge #AdvancedGPTs #D&D #AssistantsAPI #AssistantAPI #GPTTutorial #GPTs #GPT
Chapters:
00:00 - Overview
01:40 - Capabilities & Dev Experience
06:21 - User Experience
07:12 - Maintenance
08:27 - Business Opportunities
09:26 - Usecases - Наука та технологія
Production quality of a 1m+ sub youtuber, which i know you'll be soon with videos like these. Really great vid! You explained everything and even the concept of the knowledge function. Really great! subbed :D
Wow, thank you so much!! 🤩
Was hoping to find such a great comparison. Well done. Many thanks!
Thank you! Glad it was helpful! :)
What's exactly what I was lookin for. Thank you man!
dude, what a great video, already subscribed. Super clear, thanks for the info!!!!
Thank you so much!! 🤩
Great video, glad I found it and your channel! I started building with the older Completions API and was able to keep strict control of pricing. The Assistants API seems much more like a black box, since it automatically controls the whole context, where the token count can quickly add up. And unlike the Completions API, I don't believe the Assistants API returns prompt and completion token counts, which I use to get the exact usage costs my bots are generating. Hopefully they change that, because I have heard horror stories of single prompts costing multiple dollars, if you have a large context. So, for now, I am sticking with Custom GPTs which have quite a bit of functionality with Actions.
This is such a great and important point!! I've definitely spent more than I expected on some assistants queries and I worry about what it might cost if I used a huge knowledge base. Who knows what it will decide it needs to retrieve? I hope they figure out some way to make it more predictable. Thanks for watching!
Well done. I don't know what the word is, but I am the kind of learner that has to see all the details, and once that happens I feel like I become quite comfortable and I can really take decisive action. You have really provided this for me, so thank you. I think it would be really cool if you got into some more use cases of more advanced apps or businesses created using GPTs or API assistants. It's nice and all to make stickers and what not, but I would really feel like your grasp can paint a picture of what these things can really do for something far more complex and would have many moving parts to it. You mentioned different GPTs or assistance working with each other, yes I would really love to see what this stuff can really do. What do you think? 😅 Anyways, well done, keep going.
Thanks so much! Yeah with the GPT store coming next week I think I will make a video talking about some ways to make some advanced GPTs that stand out. Thanks for the ideas!
Thank you for creating this!
My pleasure! Thanks for watching! 😄
What a great video, comprehensive and well structured 👍
Thank you so much!! :) Glad it was helpful!
Thx for a great explanation. You really broke it down and showed the strengths and cool ways we can use each tool. Now I kinda get it. ^_^
Glad it was helpful! Thanks for watching!
Great Video! this really helped me figure out the next steps for my project! i was hitting a brick wall
Awesome!! Glad it helped you!!
Great video! Thank you. Very clear and concise.
Glad it was helpful!
great video
Thanks for watching!!
Great video on the differences! Any thoughts on the 40 messages/3hour limit?
For example, If I share my custom GPT application with other Plus users, then if someone uses it 40 times in an hour, then I, as the app creator, would ALSO need to wait before prompting ChatGPT again?
But if I write my own application, using assistant APIs/Secret-Key (MindStudio, AirTable), then I get around the limit but must pay for every use?
Thanks for watching! Everyone's usage is related to their account only - so others using your GPT won't count against your own cap (but you using your own GPT will). You are right that creating your own assistant would work around the rate limit, but then you'd have to pay the API token cost!
I really enjoyed this video. Very comprehensive and detailed explanations of the similarities, differences, use cases etc.
Thank you for creating this. I have a better understanding of how they work.
Thanks for watching!! Glad it helped you 😄
Super useful vid. Thank you!!
Thanks for watching! :)
Very valuable vid, Thanks !
Glad you found it helpful! 😄
So what is the difference between feeding data to an assistant in say a json file vs using embeddings and a vector db? Like say, if I want to create a discord bot that responds to users based on knowledge... which way would be the right way to go about it? or what are the main differences between these approaches?
Embeddings and a vector DB were the only way to achieve this result prior to the Assistants API being announced. Now you have the option to take either approach.
Assistants files are going to be a bit easier since you don't need to manage the vector DB, but that comes at the expense of flexibility and scale, since the amount of data and number of files is limited and you don't have control over how the data is retrieved. With Pinecone DB for example, you can combine keywords and attribute filtering along with the embedding search, so you can potentially get more accurate results. But then you need to write code to generate embeddings, store and retrieve them, etc.
I'd probably recommend starting with the assistants files and if you are hitting limitations, explore a more hands-on RAG approach (I have a separate video on this btw). For a discord bot, I'd make the Assistant via the UI, upload files there, and use the ID in my code for creating threads and runs. If you need to refresh data regularly, you could write some code to store that data into files and add it to the assistant. Will be making a video on this soon!
@@VoloBuilds Thank you for your response! I did already use your assistant discord bot actually :) however, I had a few issues; Mainly, gpt4 being too expensive and 3.5 not being good enough since it has to go through all the json data after every query which makes it slow. It also gets confused if there's several files to pull data from.
Would you consider perhaps creating a video on how to work with embeddings and a pinecone db? Ideally, I'd like to make a discord bot that would learn from the server chat data (I don't know if that falls in line with what you mentioned earlier), or at least one that works well with 3.5.
I really appreciate your responses, I'm a big fan of your channel. Maybe you also want to consider having a link up to support it.
Thanks for the support and ideas :) I am about to dive into assistant api + files for my next video and I have heard some issues with it so I'll be sure to cover that. I do have a video talking about RAG conceptually that I made before the Assistants API was announced, but it's not a hands on tutorial. I may come back to that subject and do a tutorial on it at some point in the future!
I think it does make sense to do a diy RAG setup in case you want to index/embed all your Discord messages. That would be hard to keep up to date with assistants files I think.
Any no code front-end tools that may work well for Assistants API use cases?
I'm not sure, but guessing some players who are already working on chatbots like Voiceflow will try to build on top of the assistants API but I haven't seen it yet. It's all so new! I'm actually exploring some possibilities of building something myself too
Nice video, thank you.
Thanks for watching!!
When uploading the DnD rule set, isn't there a copyright issue?
They actually make the core rules available for open and public use via the SRD which is what I uploaded! available here: dnd.wizards.com/resources/systems-reference-document
Great video! I was looking for these answers and it is super clear now. I am building a no code software with a specific use case. Using Chat Gpt was a must for me (though my own interface). This means I necessarily need to know how to code?
I haven't used many no-code platforms, but I expect you'd be able to wire them up to the Assistants API if you want your own interface but don't want to code!
thx this video explains it very well
Thanks for watching! :)
i have been trying to replicate my custom GPT which was great at anwering questions from my knowdlege base (but cant share so its useless to me) with an assistant API which supposedly updloaded the same file, used the code interpreter but the answers are totally random and it keeps insisting the info is not in the file!! are you sure they have the same knowledge cabilities?
They should use the same infrastructure under the hood though I can't prove that :) it sounds like something might have gone wrong with your upload or maybe you're pointing at the wrong assistant id? I have a video about configuring the assistants API with your own data that could potentially help: ua-cam.com/video/JzxUW0ZT4to/v-deo.html
have you actually tested it?? I have spent hours comparing the 2 based on the same knowledge text file, GPT gives perfect replies every single time, the assistant does not read the text file unless i explicitly ask it to do so repeatedly and even then it adds things like imaginary contact numbers @@VoloBuilds
Huh, that's weird - I am not sure why that would happen. When I tested it, things worked pretty well and consistent across the two. Maybe they changed something since I last played around with it? Maybe there is something buried in the prompt that is causing it to misbehave or instructions in the GPT that make it better? Not sure, just throwing ideas out there
For the usage of gpt API, do I need to pay for the usage of the api calls?
So there isn't exactly a "GPT API" - so you might be referring to 2 potential things. Assistants API, yes you need to pay for all the API calls. For Actions invoked by a custom GPT, it depends on what API you are calling (it won't be the OpenAI API) - so if you use something public, it could be free - or if you create your own custom actions, you'll need to find a way to host the API yourself so likely need to pay for hosting costs. I'll be making a video soon regarding deploying your own APIs for custom actions.
I have a test case for you. Do a medium dive on getting a GPT built with a GPT builder and share a simple action. Cut this is video # 1. Video # 2 do a simple dev chat with "Hello GPT4, what are the top five popular colors?" Cut video#2. See which video is more popular? You may find people that watch both. I will watch both. I have a 3rd case to use the Dev API, which is to write a Custom GPT where the end user is someone who does a git clone of your repo and you allow API key parameter/ variable so they get billed. Open source notoriety is better than pay. Especially if you can build a trusted dev team to grow the open source. Do a video on third. Cut video#3. I'll watch all three. Do a video on how the hell we keep up with this text sh*t? This AI is so full of rabbit holes. Down some of these holes were startups likely charging and were crushed with OpenAI over the last month. Someone needs to do a UA-cam video on all freemiums that got crushed like this.
Thanks for all the great ideas!!
Can you make a Video how to write a Chatbot with the assitent API? @11:00
Yep! I have a video about assistants API that should help you get started. Currently working on a video showing how to use files with it. Then I'll have another that covers how to deploy a chatbot to your site!
u are awesome dude!
Hey, I have seen someone saying it is not good to totally rely on openai API especially if you are an indie developer, it may stop working or shut down bcoz of the recent drama in openai leadership.....what do you think buddy?
This is a great observation and although the drama seems to have settled down, the API, models, and pricing could all change in the future so it's good to build in a way that is resilient to such changes. Thankfully there are competitors such as Anthropic that offer LLM APIs as well so I think we have options.
I expect we will see some open source version of the assistants API at some point because the abstractions are really great and can be applied generally.
The other major thing to consider and keep in mind is making sure to build stuff that goes beyond being a simple wrapper around OpenAI. I think a lot of startups that were simply feeding PDFs and files into ChatGPT are effectively out of business now with the assistants API so it's important to build stuff that solves specific problems for your target customers and does it in a way that OpenAI isn't likely to absorb into their core offering.
@@VoloBuilds thank you for the valuable points...
Just developing and running few tests with Assistant API cost $$$. Imagine if the app becomes popular… the cost is still prohibitive as many tokens are flying back and forth for simple use case. Better have strong business case.
This is such a good point. I might end up doing a video deep diving costs because it's so hard to predict right now - very much a black box and I'm seeing expenses rack up too.
is it possible to integrate a Custom GPTs to anywhere ?
You can have a custom GPT use Actions to send data elsewhere but you cannot let users access and chat with the GPT anywhere except the ChatGPT interface. You would need Assistants API to make use of OpenAI outside of ChatGPT - then you'll have a lot of flexibility!
@@VoloBuilds Can i create same thing as custom gpts with assistant api
There are differences as described in this video, but at a high level, you can use knowledge + actions and build a chat bot using the assistants API
@@VoloBuilds thanks
I'm interested in AI, but I'm completely new to it, everything sounds like a new language, dont know where to start from.
Hey, thanks for leaving this comment - I think a lot of people are in the same boat! I would love to make some content to help out people like you - but it's sometimes hard for me to think of what would be some good basic questions or subjects to cover. I'm curious, what are some of the initial questions you have or concepts you'd want to see explained?
@@VoloBuilds thanks for your interest in wanting to know how best to help me, that means a lot.
Well, I don't know exactly what to ask, but I was thinking; if one could share a story of how he/she got started with AI, that would be a solid bedrock for folks like me who are so ignorant of the subject they don't even know the right questions to start with.
You got a new follower.
Thanks for the knowledge/information you've been sharing.
Is it not very relevant that ChatGPT and GPT Agents also have this static instruction set constantly there? While with the API assistant you can start with a blank page, thereby a benefit if your good at writing instructions, and bad if you cant write good instructions/prompts. Also data from GPT is used to train the model, while with the API assistant its not.
🎯 Key Takeaways for quick navigation:
00:00 🤖 *Custom GPTs and Assistance API are powerful AI features from OpenAI.*
00:28 🧩 *Custom GPTs are built on top of Chat GPT, making it easy for users with Chat GPT Plus accounts to access.*
00:42 🚀 *Custom GPTs are user-friendly and require minimal coding.*
00:56 💡 *Assistance API allows coding for highly flexible applications with custom logic and UI.*
01:24 🔄 *Maintenance is simple for GPTs, but Assistance API requires more responsibility.*
01:37 💰 *Business opportunities may favor Assistance API due to its monetization potential.*
02:31 🧠 *Both options offer knowledge and tool access.*
03:38 📜 *Context handling is automated in both GPTs and Assistance API.*
04:07 📱 *GPTs use Chat GPT interface, limiting UI customization.*
04:34 🛠 *Assistance API provides flexibility in creating custom UIs and features.*
05:14 📊 *Assistance API offers better visibility and analytics of user interactions.*
05:54 🖼 *Image generation and vision are currently available only in GPTs.*
06:09 📂 *Assistance API allows dynamic knowledge updates, not limited to uploaded files.*
07:17 🛠 *Maintenance is simpler for GPTs, while Assistance API requires more management.*
08:40 💸 *Future revenue-sharing potential with successful GPTs, but differentiation is a challenge.*
09:09 💼 *Assistance API offers more business opportunities with potential premium models.*
09:37 🤔 *Choosing between GPTs and Assistance API depends on specific use cases.*
great content! thanks. but the hands, the hands.... :)
Skip the cheesy stock footage clips which add zero value to your video
Thanks for the input; I added them to enhance the visual storytelling of the video which helps some folks retain what is being said
I'm with volo
I'm also with Volo. Do your thing bro.... Very useful video imo.
why u mad bro
are you dancing or trying to explain a thing?
😂
cringe!!!