I made a Custom GPT this morning to review and create KZread ads. I scraped ~50 of the top KZread ads I could find, analyzed them all (used Harpa for that part), and then fed them into the Custom GPT, along with a client ad that wasn't performing very well. I adjusted the ad, based on the AI's recommendations and threw it into a split test... Control version's still at 10% making to the end of the ad AI version is at 73% making it to the end of the ad with a much higher CTR Pretty. Darn. Cool.

@capreal26

7 ай бұрын

63% CTR lift!!! This is insane. Work worth 000's of $

@itsdrutime

7 ай бұрын

Dude how do I email you to chat about this lmao

@aarongarcia2911

7 ай бұрын

Nice! Was the video and made by AI with the new AI voice as well?

@aarongarcia2911

7 ай бұрын

Thanks for letting me know about Harpa as well!

@lenghia9722

7 ай бұрын

could you elaborate or create a guide on how to scraps and analyse with harpa?

Great news and update. Still getting my head around all of this (as a non tech expert) but for the man in the street it seems massive. Thanks and keep the news coming.

Great video. Would love for you to show us how to program functions into the assistant! 🤙💯🙌

Any information on how to use functions/actions/schema? Would love to see more advanced tutorial

Highly educational! To everyone - you need to REQUEST GPT 4 API access. Yeesh YT comments. Surprising to see negative ones as well. I've already used Assistant for several tasks!

@Shykar0

7 ай бұрын

And where would one do that?

This is HUGE man !

Thanks for the video. Can we add voice at all and expect voice response? Does not seem so, which makes my next question if we can clone a voice for those responses.

"The living begeezus" lol. loved it

thank you King

Thank you.

just tried all the options in the "model" field but it does not make me save the assistant. I am using GPT 3.5

I liked this before I watched it

Very cool 😊

Hey Wes are you still following updates on the modern Turing test?

Does it send the full instructions with each request? That could account for some of the cost, if it is sending redundant instructions. Hopefully we won't be multi-billed for the instructions in the future, if it is the case currently.

👍GPT-4 provides impressive new NLP capabilities, including retrieval of knowledge from custom uploaded files. The OpenAI Playground interface allows anyone to quickly build customized AI assistants by providing instructions, choosing a model like GPT-4, and uploading relevant data These assistants can automate tedious tasks like generating KZread video descriptions and timestamps to add interactive chapters Wes gave a helpful overview of the new AI features announced at the OpenAI conference and showed how to practically apply them. Many creators are excited about the possibilities enabled by these AI advancements and are eager to start building their own assistants.

I tried to use it with the model 'GPT-3 turbo' but I got the error 'Run failed You exceeded your current quota, please check your plan and billing details.' now, I have to wait or have to pay any fees?

how come I can't select GPT 4 models even tho I have a plus subscription?

@WesRoth Great video, as always. Are you gonna make a video about GPTs as well? I mean it is basically the same but I think GPTs will be the more popular stuff than assistants. Also you can share them in a very cool way and they are free to use (for plus users)

@kevinl.9657

7 ай бұрын

It’s basically the same. GPTs are for non-programmers and Assistants are for programmers. The one used in the video is the playground for Assistants. It’s used to experiment the API without having to code. In other words, GPTs is the no-code platform and Assistants is the code platform

@hiddnhitower4620

7 ай бұрын

Do I need a team of assistant Engineers or a team of Engineer Gpts. 😅

@fabiankliebhan

7 ай бұрын

I think only the assistants can work as a team

@bestslopedesigns9429

7 ай бұрын

Dude - THANK YOU. I have been searching and scratching my head about what the heck the difference is between GPTs and Assistants. Zapier allows for integration between GPTs, as well as with Assitants. They can both access files, etc. I was really wondering. So, I'm hearing, then, that Assitants just have more "flexibility" than GPTs, right, if you can code? Thanks you so much, @@kevinl.9657 !

@bestslopedesigns9429

7 ай бұрын

@@fabiankliebhan I'd love to see a video on how to build out a team of assistants, or even a workflow map of a use case! Thanks for this question!

Legend

Do you thinks these would be good enough yet to run an enter e-commerce store?

Can you make an advanced tutorial? I would love to see the Limits

I've been hammering the assistants API and I don't seem to have any such limit. Maybe because I spend an average of $300-$500 a month on the API, I get a higher rate limit.

Nice video, I'm going to try out making an AI in playground when I get home from work, if I don't go to sleep after putting kids to bed 😂 Btw, can you also do a deep dive on making a GPT and using the actions? I don't know how to use the actions functions within GPT, and I want to know how I can use it. (I'm making an epic rap battle AI, and I need it to automatically start playing a rap song in the background 🤣)

@kevinl.9657

7 ай бұрын

Functions are for programmers. If you know how to code, just search for openai function tutorial here in youtube. For GPTs, as far as I understand, you can do the same but it’s called Actions

You are doing God's work!

In the models drop down I only see GPT-3.5 based models. Anything special we need to do to get the GPT-4 based ones to show up like in your example?

@TTalkWithTaylor

7 ай бұрын

You need to have a paid account. It’s like $20/mo.

@samuelflippin1890

7 ай бұрын

That isn't correct. At one point you had to request and get access to the GPT-4 API. I'm not sure if you still have to do that. Perhaps you have to have spent some money already but you do not need to have the 20/mo subscription.

@meathunderify

7 ай бұрын

OR it's possible you haven't put $$$ into the API after some particular date ... I forget which... In the very early days I put in ten bucks and began to get irritated bc I was never getting access to GPT4's API... searched their Forums/faq... a guy had the same problem... and it was instant access after I added more $$$. That *could* be your problem

@jonatan01i

7 ай бұрын

@@samuelflippin1890 Copied from X """ Sam Altman @sama GPTs are now live for all ChatGPT+ subscribers! """ This pretty much feels like yes you need the subscription.

@kevinl.9657

7 ай бұрын

You need to have access for GPT-4. As I remember, you need to spend at least a dollar first then you’ll have access. Try to use the playground first until you spend a dollar

Nice, someone else using an agent designer :) like functional programming your agents should be. - Yoda

0:00 chaotic neutral or bust! XD

I have no idea what this is for, but can it do interesting things? What can it code?

what is the difference between GPTS and assistants

What do you think of my custom GPT for Public Relations?

How does it connect to an app with a UIX?

how much was your herman miller?

OpenAi Assistant says you can upload up to 20 * 512Mb knowledge based files - does anyone know the cost implications?.. I presume the data in the knowledge based files isn't part of your tokens - but how is the pricing worked out?

how can we generate images with this dalle is not in dropdown list

nice

I love this. But why does this video have any timestamps?

@WesRoth

7 ай бұрын

lol, that was the plan originally. will add it today.

What API assistant can do that GPTs can't?

🎯 Key Takeaways for quick navigation: 00:00 🚀 Building Your Own Assistant - Creating your own assistant on OpenAI Playground. - Selecting assistant, naming, and providing role instructions. - Adding timestamped transcripts for the assistant to analyze. 02:08 🔄 Model and Retrieval - Introducing GPT-4 Turbo and its capabilities. - Discussing the importance of retrieval for accessing knowledge from uploaded files. - Explaining the retrieval augmented generation concept. 03:46 📂 Uploading Files and Code Interpretation - Uploading files to the assistant, including the timestamp document. - Discussing the code interpretation and functions feature. - Demonstrating the use of the files section for general project resources. 04:00 🤖 Running the Assistant - Saving the assistant and initiating the interaction. - Providing a sample timestamped transcript from a KZread video. - Showcasing the summary and section breakdown generated by the assistant. 06:29 💳 Token Limit and Continuing - Explaining token limits and the rate limit challenge. - Demonstrating the process of continuing the interaction despite token limitations. - Emphasizing the potential cost-effectiveness of using the assistant. 08:22 🌐 Broad Applications and Cost Analysis - Expressing the time-saving benefits of automating tasks with the assistant. - Discussing the vast potential for creating various assistants tailored to specific needs. - Providing a cost analysis example based on token usage during the demonstration. Made with HARPA AI

For me the data you can upload for retrieval is too limited. For most things interesting to me I need a much larger knowledge base. So stuck with the regular chat API and my own vector database.

@DJ.Generation

7 ай бұрын

Yeah i came up with something called "promot software" to convert the chat bot into like a interactive task specific software. I swear OpenAi looked at mines and other chats for the GPT idea. But regardless. It is basically a elaborate "prompt code" with natural language instead of coding But its like 30 K- 50 K+ characters, but the GPT can only handle 8K. So its still better to use ChatGPT general xhat since they made the updates. It functions just fine in there. I just cant sell them on the GPT store for now. So for now i think all innovators still jave the leg up.

@avi7278

7 ай бұрын

I think the limits are by design. OpenAI wants you to use assistants like you might use a function in a codebase. SRP (single responsibility principle). It makes them more money than if you're able to get everything you want in a single call. Also imagine the sheer amount of data storage and reources. If everyone is allowed to upload 1000 files and have them all vectorized (to my knowledge we are not billed for this). Perhaps they will open this up to pay to play later.

Can you have another Output than Text? Can you create an API that sends this tool E-Mails and sends answers Back? Just a newbie here

@kevinl.9657

7 ай бұрын

If you are not a programmer then I think you can do that with Zapier’s plugin. Or when the time comes, Zapier, I think, will have their own GPT in the GPT store. Otherwise, learn how to code and do it yourself. That could be fun.

@baumwollejr

7 ай бұрын

@@kevinl.9657 thank you!

Can you make a video on how to put custom actions with APIs. I put one in and it completely broke my bot.

@apmanti12

7 ай бұрын

Yes, i need info on how to use those too

@BGIANAKy

7 ай бұрын

@@apmanti12 so I think I finally figured it out. And ChatGPT updated its backend it looks like. I was doing google my business listings so I asked ChatGPT to guide me through doing it that setup, doing scopes and all of that. Then I ran into the problems yesterday. Today though, the actions were different and after using a json validator tool, ChatGPT and a few different scripts it finally took it. Just to let you know, I had to go on OpenAI’s article where they talked about allowing custom APIs and I had to tell the builder it was fine to get it from telling me it was a simulation. Anyways, now I need to figure out how to authenticate easily with ChatGPT and gmb.

@Mezzers83

7 ай бұрын

I was playing around with this last night and there's a limit to what it can consume. See if you can retrieve smaller datasets for your actions

i tried all the options in the "model" field but it does not show me the GPT4,only have gpt 3.5.does it need the gpt plus modle only?but i am the gpt plus user,why it not show me the GPT 4 modle?need i buy API first before i use this function?i mean Playground Assitant . please reply me,thanks so much!

@louis-philippemaurice1607

7 ай бұрын

You need to add money to your account. You can't do it with the free credits.

@lucy-917

7 ай бұрын

thanks for your reply,i have chatgpt plus,what i wonder is need i pay additional fee for the playground assistant?@@louis-philippemaurice1607

At this rate GPT4 is doing work which would correspond to you hiring an assistant at a rate of $1/hr. Oh and it does the work instantaneously. I'll leave it up to the reader to think about all the ways in which this will impact the labor market lmao

Why not write a function to get the transcript programmatically, then use 3.5 turbo to create the summary? I’m not being negative for negatives sake, this new way of doing it just looks less efficient and more costly than functionality we already had access to.

@WesRoth

7 ай бұрын

Retrieval, I think, is only available on the new GPT model. Also, I think you would bump into the context window on longer videos. Don't get me wrong, it's possible, I was just trying to showcase what was possible with assistants. There will be cool use cases that won't be 100% efficient and better to do with 3.5, I agree.

@hansofmadata3565

7 ай бұрын

@@WesRoth I really appreciate you and other KZreadrs who take the time to demo new capabilities. It’s a really engaging way to keep up to date without having to read Medium articles or documentation.😁 My feedback isn’t a reflection on your specific example, or your coding capabilities, just an expression of my frustration at these new features. As a dev who uses OpenAIs LLM I’m more concerned with how OpenAI will scale and where their future focus will be if the platform becomes flooded by non developers using ChatGPT as a no code tool to create endless copies of QA chat bots trying to be popular on a monitized App Store. Imagine the endless amounts of compute that’ll be used to store copies upon copies of Andrew Huberman podcast transcripts. A dark future indeed 😂. Thanks again for your hard work Wes!

How about chaotic good? I do not want to be left out. ;)

The timestamp identifiers are a bit long when navigating on the video player (all end in three dots ...) and for ease of use (reading takes more time than watching the video). Need to be more concise. Thank you for the upload.

Makes a video about a new YT description generator… doesn’t use it to generate a description for the video though 🤔

@WesRoth

7 ай бұрын

haha! yeah, YT was taking too long with the transcript lol. I have a workaround, I'll be using it going forward

How can I export an custom GPT to a website?

@qa1e2r4

7 ай бұрын

They did an example during the reveal look it up

@azj77

7 ай бұрын

@@qa1e2r4 I can't find it would you be able to send me a link?

how is that 30k tokens??

🎯 Key Takeaways for quick navigation: 00:00 🤖 *OpenAI GPTs are now available for all Chad GPT plus subscribers, and you can create your own GPT assistants.* 00:29 📝 *To build your own GPT assistant, start by going to the OpenAI Playground, select "Assistant," name your assistant, and provide clear instructions for its role.* 02:21 📚 *GPT assistants can use retrieval to access knowledge from uploaded files, making them capable of remembering information.* 04:14 📑 *You can easily create structured summaries and section breakdowns of KZread videos by uploading their transcripts, saving time and effort.* 08:36 💡 *Creating GPT assistants can automate tasks that were previously time-consuming and tedious, saving time and money in the long run.* Made with HARPA AI

Cool! Wouldn't it be better if you'd asked the assistant itself to use bing, go to the video's URL, download the captions and then go through with it?

@ensane

7 ай бұрын

would have to be prior to april 2023

@reza2kn

7 ай бұрын

@@ensane Nope, it can browse the web for you.

@ensane

7 ай бұрын

@@reza2kn 😧

@aarongarcia2911

7 ай бұрын

I think if you use Google bard it is up to date. So maybe use bard. I never thought to do this. This is genius

first I get my jaw from the floor. Then I will go to sleep to get some ideas. Then I wake up and make myslef nice cup of tea Then I will use that for sure! :P

@WesRoth

7 ай бұрын

Perfect!

Accessibility, accessibility, accessibility. A lot of developers and website creators do not know how to incorporate accessibility into their products. A lot of visual media out there is not described. And with XR interfaces becoming the thing I am afraid the world is going to become more inaccessible.

Instead of getting the response in a chat, why not just ask it to return the response in .txt file. this will save you some tokens i think

@sirvladislav

7 ай бұрын

It would cost the same. It doesn't matter where you will save the output

🎯 Key Takeaways for quick navigation: 00:00 🎄 Introduction of GPTs for all Chat GPT Plus subscribers and creation of personalized assistants. 00:15 🛠️ Tutorial on building a custom assistant in five minutes using OpenAI's platform. 00:29 🔍 Importance of selecting 'assistants' in OpenAI's playground for specific tasks. 00:43 ✍️ Setting up an assistant for creating KZread video descriptions with detailed instructions. 00:58 🤖 Highlighting the need for clear communication with AI assistants. 01:11 🎞️ Instructions for segmenting KZread videos into chapters using timestamps. 01:39 📑 Providing format examples for the AI to follow in video description creation. 01:53 🧠 Selection of the GPT-4 model for the assistant's functionality. 02:08 🗃️ Explanation of the retrieval feature for accessing uploaded files. 02:36 ⚙️ Overview of additional functionalities like code interpreter in the assistant setup. 03:04 📁 Discussing the purpose of the file upload section in the assistant's interface. 03:18 📋 Process for obtaining and uploading a timestamped transcript of a KZread video. 03:46 💡 Highlighting the ease of creating KZread video descriptions using AI. 04:00 🚀 Launching the assistant with uploaded file for generating a video description. 04:14 📊 Assistant's successful generation of a detailed and accurate video summary. 04:53 🎨 Discussing potential for further customization and refinement of the assistant. 05:07 📌 Accurate time-stamped section breakdown provided by the assistant. 05:47 📈 Addressing the token limit in the AI model and its impact on output. 06:16 ⏳ Managing token limitations for continued assistant operation. 06:44 ✅ Completion of the video description task after managing token usage. 07:12 🕒 Highlighting the time-saving benefits of using AI for KZread descriptions. 07:40 🌐 Emphasizing the vast potential and daily use of AI assistants. 07:54 💸 Discussion on the cost-effectiveness of using AI for automation tasks. 08:08 💰 Estimation of token usage and cost for the assistant's operation. 08:22 🔄 Exploring further optimization and functionality of the assistant. 08:36 📢 Encouragement to utilize and explore AI tools for efficiency and innovation. Made with HARPA AI

GPT4 vanilla can do all these too. Just try it. . I think these agents with function calling is better suited for making interactable AI characters in video games, than anything else.

I think we will come to a point where AI itself will create its own gpts 😅

Ive been watching your videos for a week, like 10 of them so far and I realized Ive never subscribed lol. Well fuck me, now I'm going on a liking rampage on your channel agghhn!!! Actually I'll have my assistant do it NVM. ;) 😉

@WesRoth

6 ай бұрын

nice lol!

Dark mode please

I never would have noticed the notations are way too long if you didn't mention it was done by a bot haha

But the Assistant API is not useable for production because the cost of tokens is unpredictable. They calculate some kind of mystirious "Context Tokens", which nobody even their customer service understand. The real token cost is about x10 of the price list!!!!!!!!!!!!!!!!!!!!!!

You didnt use the Assistant on this yt video 😂

Being overly polite in the prompts is actually a bad idea. Sometimes GPT4 will interpret "please" as a flag that means optional behavior and then will pick and choose which instructions to follow. Do not use "please". Do not use phrases like "would you kindly".

Hi English is not my native language . what does this reference to Jesus means specially the "B" in your expression: "there's no doubt in my mind that I'm going to be using the living B Jesus out of this on a daily basis " Thank you in advance

Hey. I watch your videos at night, sometimes in bed. Anyway, I can't as it's so bright. Have you considers dark mode ? You are literally the brightest channel in my list. Can't watch it.

@QazDiaries

7 ай бұрын

😂same

@aarongarcia2911

7 ай бұрын

Do you not know how to dim your screen? Or put on nightlight on your device?

@shamrockisland

7 ай бұрын

@@aarongarcia2911 do you not know how to reply without making assumptions? Well we both know by your reply you don't. So I will correct you like a child. The device is already fully adjusted, wait for it, that's why I posted my question, maybe consider not replying further. 😂😂😂

@aarongarcia2911

7 ай бұрын

@@shamrockisland it was meant to make you feel like a child though. Because he won't adjust just for you. Just dim your screen or use the nightlight version. You're one person and it's a video for many

@shamrockisland

7 ай бұрын

@@aarongarcia2911 deflecting much. Again with the assumptions. So I refer to your remarks as childish and you attempt to deflect it back onto me lol. I haven't said anything childish. As for he won't ? Are you making further assumptions. Lol deplorable sign of your obvious inherited stupidity. He's doing this for a living he will take educated posts onboard. Can I please ask you to stop replying as you're beneath me. Thanks.

ugh

God do I have to keep commenting. Your audio hurts to listen to. Like nails scraping metal. You need a pop filter or something so the S sound and D sound aren't so loud compared to the rest of your voice

@conall5434

7 ай бұрын

I think thats just you bro, sounds fine to me

A KZread description creator does in no way indicate the power of this new tool. If you can't find something actually interesting or useful to non creators why make the video? I mean it leaves me feeling like this is the coolest GPT out there, which makes me sad that I'm following this so closely.

@jaysonp9426

7 ай бұрын

At least programmatically download the transcript and have a webhook for when you post a video or when you drop a video into a folder.

@samvarcoe

7 ай бұрын

It's a simple use case but these tools have only been available for a couple of days. This example is very niche but abstract what he's doing and apply it in general. I think Wes summarised the value really well when he pointed out that, with a very small amount of effort and time, he was able to automate a task that usually takes him 20mins for a cost of about 25 cents. If you're a knowledge worker making more than 75 cents an hour then I'm sure there are plenty of ways you could leverage this technique.

@uploaderfourteen

7 ай бұрын

Criticises the video for not in any way indicating the power of this new tool…. In doing so proceeds to not in any way give any examples of the power of this new tool. Makes me sad I am following this so closely lol.

@primeryai

7 ай бұрын

It seems that you don't understand what was demonstrated here. The use case isn't the point

@qAidleX

7 ай бұрын

He is teaching you math. The basics are 2+2. 2+2 is boring and non-useful, but the capability of the mathematic systems behind 2+2 is profound, ya know?