If people have repos/tasks that you want me to try in part 2 post them here and I will record the video in about 12 hours from now. Please give specific examples etc.
@@WatchNoahhe pointed out in an earlier video that the temp slider is locked for Gemini 1.5 and that he doubts they're actually using a temperature of two.
I want to look at the prisma repository. I would ask for: " A step by step implementation plan to allow prisma compile and introspection to create and compile each table in its own folder within a designated schema directory as per the schema config, while also maintaining type safety between the partial schemas." This is a years old feature request on github with over 300 comments, overwhelming support, and a complete lack of developer attention.
MARVELOUS PRESENTATION! I really, really appreciate that you make sure it's easy for people to follow what you're doing. I think many TH-camrs pay a lot more attention to having the "right SHOCKING caption" or a "thumbnail with their mouth and eyes open surprised as hell" than they do about integral stuff like making sure you're getting the message across.
Sam can you please tell how you get to that "Google AI Studio" interface that gives you options to insert files, folders etc. My Vertex AI studio doesn't have those options and only allows specific files formats for uploads (mostly multimedia only). I can't seem to find any tutorial on that either. Thank you !
@@samwitteveenai Thanks Sam. I love your videos. It's the perfect blend of concepts and ideas with actual code implementation, at the right level of detail, focus on important points, and even the speed of content delivery. Just outstanding !
This 'program' is still broken. When you can ask it questions without going PC on you and telling you it cant do this yet because Im still learning is BS... stop charging until you (fix) this shit!
Summary of books and generating code and code analysis from codebases are the most interesting things. Thanks for exciting video Sam, looking forward to part 2!
Hey @samwitteveenai, I don't want to sound like a noob however how the heck you could provide repo files to ai studio? Currently, I can't see any way to upload my code to it.
That was very interesting, thank you. It would be nice to see some copilot vs gemini comparisons. Not sure about code whisperer, that seems to be set up for AWS integration, but might be worthy of inclusion. Thanks again.
I'd really like to see this working for the FastAPI codebase, I've got a bunch of students learning it right now and would be curious how something like this would help them. For an example, get it to write a FastAPI api for a library application. Bonus points for making it generate a react frontend :) good idea for a vid, thanks for the crew AI info too.
Yeah I don't understand why this guy is doing such limited and timid use cases he has a beast at his disposal. How far can it go in created your standard Saas style full stack application?
Do you have a sense for the output token limits? GPT4 limits output at 4,096 tokens, which I've found restrictive for certain use cases like writing out long code scripts. Curious if the output token limit for Gemini Pro 1.5 is higher.
not sure they want me to say currently (and the model may change before wide release) I don't think you should expect super long output though. I have found it to be similar to a lot of models in that it is hard to get it to output something too long in a single output.
@@samwitteveenai Thank you and appreciate your answer! The way I usually test this is by providing a very long code script in the prompt and asking it to (i) change a few things and (ii) send the entire code repo back to me with the changes. Typically, the model starts rewriting code script and then typically stops at around 4k tokens. If you have a second to try that method out I would highly appreciate it. 🙌
Gemini refuses to answer questions about code in C++ that it deems unsafe - like copying memory fast, or refused to provide code examples when the user is underage.
Summary of Key Points: 📌 The video explores the use of Gemini 1.5 Pro for various tasks related to coding, focusing on analyzing the Crew AI repository to generate examples based on the code. 🧐 Gemini 1.5 Pro was used to set up a scenario where a simple bot with two agents, a customer and a salesperson, interacted to discuss air conditioner upgrades, showcasing the capabilities of the AI framework. 🚀 The video demonstrates the transition from using OpenAI to Gemini Pro for language model generation, highlighting the differences in outputs and the need for specific package installations. Additional Insights and Observations: 💬 The AI-generated code showcased the ability to create dialogues between agents, handle tasks, and provide outputs based on the given scenarios. 📊 The video emphasizes the potential of AI tools like Gemini Pro in understanding and analyzing codebases, predicting future improvements in such technologies. Concluding Remarks: The video illustrates the practical applications of Gemini 1.5 Pro in coding tasks, showcasing its ability to generate code, handle scenarios, and potentially enhance codebase understanding. The transition from OpenAI to Gemini Pro demonstrates the evolving landscape of AI tools in the coding domain. Generated using TalkBud
So it sends the whole codebase with each prompt there I guess? No embeddings, right? So, 40k tokens is like 200k characters. With pricing of Gemini Pro (and 1.5 is probably going to be more expensive), first request you send will cost you around $0,05 (and each next request in the conversation will be more expensive than the previous one). It's gonna stack up pretty fast, considering you will have to ask multiple times to get stuff done right. That is if you are in European Union, as the rest of the world probably can use it for free right now.
Is this available yet? When will they release it to the general public because I don't want to pay for the Pro until this is available because this seems useful. Unlike most of the other stuff Google makes.
I don't think it is. I just tried to repeat these steps and got a laughable lack of results. It says it can't even write much code for me too, which is odd?!
If people have repos/tasks that you want me to try in part 2 post them here and I will record the video in about 12 hours from now. Please give specific examples etc.
Can you use AutoGPT, SuperAGI, GPT Pilot or GPT Engineer
I think you should put the temperature down. temperature=2 seems quite a lot for coding tasks
@@WatchNoahhe pointed out in an earlier video that the temp slider is locked for Gemini 1.5 and that he doubts they're actually using a temperature of two.
I want to look at the prisma repository. I would ask for: " A step by step implementation plan to allow prisma compile and introspection to create and compile each table in its own folder within a designated schema directory as per the schema config, while also maintaining type safety between the partial schemas."
This is a years old feature request on github with over 300 comments, overwhelming support, and a complete lack of developer attention.
It shouldnt be an AI-related repo, because otherwise most people cant check if Gemini is answering correctly
Both gemini 1.5 and CrewAI are very popular, and the youtuber is very smart. The two are combined together. Thank you for your careful production.
MARVELOUS PRESENTATION! I really, really appreciate that you make sure it's easy for people to follow what you're doing. I think many TH-camrs pay a lot more attention to having the "right SHOCKING caption" or a "thumbnail with their mouth and eyes open surprised as hell" than they do about integral stuff like making sure you're getting the message across.
Was just about to shoot a thumbnail with my mouth open and you have saved me. Thank you!!
Hahah. Can you please try generate a full stack app asking it to provide the front end in next.js the react framework?
Keep Going!.
Awesome video. Thank you! Can you share the Notebooks code to test. Thanks
Sure here you go colab.research.google.com/drive/1VACRSn9LqbFcJlPe4qtryNY_3Jl4ssqS?usp=sharing
Your amazing. Thank you @@samwitteveenai
Sam can you please tell how you get to that "Google AI Studio" interface that gives you options to insert files, folders etc. My Vertex AI studio doesn't have those options and only allows specific files formats for uploads (mostly multimedia only). I can't seem to find any tutorial on that either. Thank you !
just make sure you select the 1.5 model. Take a look at the new vid I just released I mention it there.Vertex is different than AI Studio btw
@@samwitteveenai Thanks Sam. I love your videos. It's the perfect blend of concepts and ideas with actual code implementation, at the right level of detail, focus on important points, and even the speed of content delivery. Just outstanding !
Where can we access Gemini 1.5?
This 'program' is still broken. When you can ask it questions without going PC on you and telling you it cant do this yet because Im still learning is BS... stop charging until you (fix) this shit!
Summary of books and generating code and code analysis from codebases are the most interesting things.
Thanks for exciting video Sam, looking forward to part 2!
Very useful - when I code with GPT4 - I have a problem with new libraries and updates. Gemini 1.5 Pro with huge context window may be the solution:)
Hey @samwitteveenai, I don't want to sound like a noob however how the heck you could provide repo files to ai studio? Currently, I can't see any way to upload my code to it.
How can reg Gemini 1,5. I still 1.0
I'm confused you said this is Gemini 1.0 in the video versus the title?
CrewAI is really cool and I love the proyect but If you want to really test context and code base you should try it with mentant
Can you provide the colab note book pls?
Sure here you go colab.research.google.com/drive/1VACRSn9LqbFcJlPe4qtryNY_3Jl4ssqS?usp=sharing
That was very interesting, thank you. It would be nice to see some copilot vs gemini comparisons. Not sure about code whisperer, that seems to be set up for AWS integration, but might be worthy of inclusion. Thanks again.
I'd really like to see this working for the FastAPI codebase, I've got a bunch of students learning it right now and would be curious how something like this would help them. For an example, get it to write a FastAPI api for a library application. Bonus points for making it generate a react frontend :) good idea for a vid, thanks for the crew AI info too.
Yeah I don't understand why this guy is doing such limited and timid use cases he has a beast at his disposal. How far can it go in created your standard Saas style full stack application?
How to get access to Gemini 1.5 pro? I only see gemini Pro in my account.
You need to apply for it and get access to the Private Preview. I just got access 4 days ago but I think I applied a month ago I think.
Do you have a sense for the output token limits? GPT4 limits output at 4,096 tokens, which I've found restrictive for certain use cases like writing out long code scripts. Curious if the output token limit for Gemini Pro 1.5 is higher.
not sure they want me to say currently (and the model may change before wide release) I don't think you should expect super long output though. I have found it to be similar to a lot of models in that it is hard to get it to output something too long in a single output.
@@samwitteveenai Thank you and appreciate your answer! The way I usually test this is by providing a very long code script in the prompt and asking it to (i) change a few things and (ii) send the entire code repo back to me with the changes. Typically, the model starts rewriting code script and then typically stops at around 4k tokens. If you have a second to try that method out I would highly appreciate it. 🙌
Maybe you should mention that Google AI Studio is still unavailable in EU.
have you tried a VPN? for AI Studio?
YOu can use it by VPN, i just got access by connecting to turkey
interesting to know.
but did you get 1.5 ?@@cybertruck2008
Gemini refuses to answer questions about code in C++ that it deems unsafe - like copying memory fast, or refused to provide code examples when the user is underage.
Seems we cannot use Gemini Pro 1.5 in a program yet, only in their playground. 😰
Try making a bin Unix package from url 😂
does duckduckgo provide search api without api key?
yes with that package at least. Thats why I usually use that for the demos.
Very interesting...
Mine made snake 😂
Wow... impressive actually...
Great video ...
Keep pushing 1.5 pro.
Summary of Key Points:
📌 The video explores the use of Gemini 1.5 Pro for various tasks related to coding, focusing on analyzing the Crew AI repository to generate examples based on the code.
🧐 Gemini 1.5 Pro was used to set up a scenario where a simple bot with two agents, a customer and a salesperson, interacted to discuss air conditioner upgrades, showcasing the capabilities of the AI framework.
🚀 The video demonstrates the transition from using OpenAI to Gemini Pro for language model generation, highlighting the differences in outputs and the need for specific package installations.
Additional Insights and Observations:
💬 The AI-generated code showcased the ability to create dialogues between agents, handle tasks, and provide outputs based on the given scenarios.
📊 The video emphasizes the potential of AI tools like Gemini Pro in understanding and analyzing codebases, predicting future improvements in such technologies.
Concluding Remarks:
The video illustrates the practical applications of Gemini 1.5 Pro in coding tasks, showcasing its ability to generate code, handle scenarios, and potentially enhance codebase understanding. The transition from OpenAI to Gemini Pro demonstrates the evolving landscape of AI tools in the coding domain.
Generated using TalkBud
❤🎉
0:00
So it sends the whole codebase with each prompt there I guess? No embeddings, right? So, 40k tokens is like 200k characters. With pricing of Gemini Pro (and 1.5 is probably going to be more expensive), first request you send will cost you around $0,05 (and each next request in the conversation will be more expensive than the previous one). It's gonna stack up pretty fast, considering you will have to ask multiple times to get stuff done right. That is if you are in European Union, as the rest of the world probably can use it for free right now.
1.5 pro is going to be cheaper than1.0 pro
Openai is over
Not just yet, they have some nice models coming as well.
"As a top-performing salesperson with an extensive knowledge of..." is certainly what a top-performing salesperson (trained by OpenAI) would say.
which is better for coding and programming tasks gpt4 or gemini 1.5 ?
Is this available yet? When will they release it to the general public because I don't want to pay for the Pro until this is available because this seems useful. Unlike most of the other stuff Google makes.
I don't think it is. I just tried to repeat these steps and got a laughable lack of results. It says it can't even write much code for me too, which is odd?!
chatgpt 4. i will stop pay them now