The SIMPLE Way to Build Full Stack AI Apps (Tutorial)
Вставка
- Опубліковано 12 лют 2024
- Gradient makes building full-stack AI applications with RAG, summarization, and sentiment analysis more straightforward. I'll show you how to use Gradient's new Accelerator Blocks feature so you can get started building AI apps quickly.
Sign up for Gradient and get $5 in free credits today: grdt.ai/mberman
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewberman.com
Need AI Consulting? ✅
forwardfuture.ai/
Rent a GPU (MassedCompute) 🚀
bit.ly/matthew-berman-youtube
USE CODE "MatthewBerman" for 50% discount
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
Media/Sponsorship Inquiries 📈
bit.ly/44TC45V - Наука та технологія
People seem to have a lot of opinions about how I should run my channel. I appreciate the feedback and I will try to use less click bait titles, but this can’t be your first time on UA-cam, right?
Please understand how much time and effort I put into make free content before you comment.
❤
Matthew, your fabulous quality separates you head and shoulders from 99.5% of what typically passes for UA-cam "content". I like to think that the audience you attract to this channel appreciates this as well, which makes truth in headlines yet another quality feature rather than a liability!
People shouldnt hate the man for trying to get his bags. Ignore the videos you dont like and watch the ones that arent like this.
Matt, thanks for all your great content. And also your investment in LM Studio!!!
I 'disliked' this vid just because it seems a little too much of a 'product placement' for me. i'm not likely to use this, but it is good to still know it exists, so thanks for the coverage nonetheless.
Thanks again for your work. It's much valued!
You do great video and are a great resource. I automagically post each of your videos to my team. I think if you keep the information quality high you'll be fine. Just don't get sucked by cheap UA-cam tricks like begging for subs, adding comments, pushing stuff to the end. It is almost cliche at this point. I'll end by saying keep up the good work and ask if you are somehow related to Nathanial Berman...because you guys look like brothers.
I do appreciate the quality content. I hardly comment but it's been one of those days where I keep seeing the same titles and I just wanted to give you advice you didn't ask for lmao all love though Matthew. You rock!
What's the local alternative? I'm not interested in cloud solutions.
Google alternative to "x" -- lots of good sites. add "open source" to it also.
I hear that. If I could get inference speeds from local models that I get from openai API, I'd be in business.
The shocking title trend needs to die.
Ok
Unless you’re seeing a sudden drop in views, don’t stop your style. You stand out among the rest for a reason. You have to know when to ignore the haters. You can’t please everyone!
You killed it?! MURDERER! it's the internet, no matter what you change someone will have something to say about one thing or another lol Thanks for having the patience to deal with everyone, I learned a bunch from you so far :)@@matthew_berman
Thank you for having a "$5" in the description. It's extremely important to me to know which videos are pure marketing. I'll still play them because I want to support you, but I almost unsubscribed a few months ago when your earlier such videos were not obvious. I think you could improve, however, with making it more obvious. All that said, I love this channel, your work, and how you present information. Keep up the could work.
I've always said (100% of the time) when a video is sponsored.
Thank you for the kind words though.
@@matthew_bermanquietly and quickly. It should be labeled at the start of every video. Not like they aren't all sponsored anyway
"And it shocked the industry!" 😂😂😂😂😂😂
no! the ENTIRE industry!
Wes is seeing his sub count dropping every time he uses "shocking"
@@YouLoveMrFriendlysame with Wolfe, they all surrendered to clickbait hype titles, unless something changes they are getting replaced soon with new blood.
I don't understand how this sponsored video is supposed to help me. There are countless tutorials available for accomplishing this using existing APIs, especially as you mentioned, as well. This includes even self-hosted options. The real question is: how many of us receive more than one request similar to these scenarios more frequently than once a minute?
Matthew, please don’t get me wrong: I love your content, I can understand sponsorship, but the more fingers come into the play the more expensive it will be.
Exactly, you you the time to intelligently express what I actually meant by saying there's absolutely no reason to use this service. It's "free" sure, probably heavily rate limited and what do you actually save? The time it takes to write a prompt to ask for a summary of the text? Just silly. There are APIs that allow you to access Mistral at .50 for 2 million tokens. You'd have to be stupid to build on top of this "accelerator", I mean "proxied API request with a prompt in the middle that you saved two seconds writing"
well said. I choose to bump this comment as an expression of my frustration and confusion.
Chris at Gradient here - we actually built models designed to do this at higher accuracy than OSS models - so it's not just an API layer. We should have clarified this on our website but it is designed specifically to perform better for enterprise workloads
@@ccubed510 that's interesting to know.
however, the frustration is not directed at the tool, but the creator. I for one, only follow him for local hosted tutorials and want nothing to do with anything hosted off my own equipment. His tutorials are of good quality, so it was frustrating to see this sponsor video pop up, in a click baity fashion, when, again speaking for myself, was expecting more of his usual local model based content.
Again, thank you for the insight.
@@ccubed510 Chris, finding the right customers must be challenging for you. The market is vast, and products seem hard to differentiate. Perhaps specific jargon needs to be developed and filled with meanings to ultimately present everything in a pitch.
Whether it's because "the best marketing does nothing if the target audience isn't in the mood," or due to the fact that B2B on this channel comes at the expense of the fan base, which would have expected a presentation/testing of a model, and is only reachable indirectly, I wasn't able to understand it.
Whether it's layers, wrappers, or a sophisticated model, show us a video that demonstrates the functionalities "works as intended." If necessary, with 1 million requests at an enterprise level. I'd love to watch the video and look forward to the link.
2:40 It's important to verify that the output of an LLM is correct. In this case, the summary it gave is not. None of those things happened in the first chapter of the first Harry Potter book. That chapter is all about Harry before he entered Hogwarts. Summarization like this isn't really what RAG is for, though. A RAG query prompt needs to match the meaning of a chunk in the document to return something relevant. A prompt that asks for a summary doesn't really fit with that approach because it will look for an existing summary within the document. The LLM presumably knows about Harry Potter from its training data without needing to do RAG and so just tried to answer based on that, and got it wrong. However, the summary given later in the video at 4:33 using the Document Summary function is accurate.
This could actually make the programming assisting meaningful. But I wonder if this kind of a system is available locally?
What is the best clip interrogator to output precise image tags/tokenS. Is there an SAAS service that simplifies fine tuning clip extraction with my custom captions/concepts (ie softbox lighting)… have you covered this somewhere? I’m working with Stable Diffusion and can deal with Python but would prefer an online enterprise solution
Matthew, don't pay too much attention to the negative backdraft from a few recent titles. The internet can act like a pretty mad hive mind sometimes. People forget that there's a human reading the comments. Anyway, hope you are well, and as always thanks very much for the AI content 👍
Much appreciated
I found the video very helpful actually. its nice to know how product's even paid ones work. How else would i find out so quickly. keep up anything you are doing Matthew
How is this different from creating your own GPT in openai? You can upload whichever documents you want and probe it then
RAG died, this is a sponsored vid.
Chris at Gradient here - it's different in that we purpose built LLMs for these tasks so they would be performant in an enterprise workload environment. Thanks!
I am SHOCKED.
As usual, FANTASTIC video! 😍 Thanks Matthew, looking forward to playing around with Gradient AI. I used your referral link. 👍
Thanks I enjoy all the content you take time to show us. Keep up the good work. Terence, Nelson , New Zealand
Great video. Very interesting indeed. Gradient is a super solid product. I’m caught trying to rationalize AutoGen, MemGPT, and RAG. What’s the best multi agent / data retrieval solution? The lake of options continues to grow.
There’s actually a new ‘os-copilot’ that controls your system and has a semantic memory feature. But, you know. Efficiency and OCD don’t play nice.
Great video but I don't know how it the title is building full stack application so I don't see any full stack related things in this video or did I miss anything please let me know?
SHOCKED! - I like your channel but perhaps save the capitalized superlatives for something actually groundbreaking.
Ok
Thanks for the great videos! Data governance question. This platform seems great, but I am wondering about data security and how gradient uses or keeps tokened or uploaded data? I have been using privateGPT because of analyzing private data and data sharing concerns. Is gradient safe in this regards?
Chris at Gradient here - the platform is designed for enterprise usage so we isolate all the user data (We're HIPAA compliant and SOC2 certified as well).
Maybe it is user error but if you use their rag, you can check in your files any time you like, but they can never leave. Beware.
we're working on enabling deletions - will be fixed in a week or two!
Amen! @@chrischang7870
I have a dumb question. My initial assumption is that this shouldn't be used with information that is meant to be confidential, right?
Summary: The video introduces Gradient's launch of Accelerator Blocks, an innovative tool designed to simplify the creation of complex AI workflows for engineers. With these blocks, users can effortlessly combine different functionalities like personalization, sentiment analysis, Q&A, document summarization, and entity extraction into their projects. This development democratizes AI application, making it accessible and free to use for anyone interested in leveraging advanced AI capabilities without extensive technical know-how.
I'm sad, you have promised more videos of other things that you have done with agents and nothing, you only post new things, which I like but I would like it if you promised to extend the functionalities of things that you show us, it would be great if you did, excuse me, I know that It's your time and I respect it, but it leaves me excited to know how a professional like you can get the most out of these tools.
I have many videos about Autogen, and more coming about crewai
Anything about the 'Create A Personalized AI Chatbot with Chat With RTX' from Nvidia?
🎯 Key Takeaways for quick navigation:
00:00 🛠️ *Gradient launched accelerator blocks to simplify building complex AI workflows without much effort, offering flexibility to combine different blocks for various workflows.*
00:26 🆓 *Gradient's accelerator blocks, including personalization, sentiment analysis, Q&A, doc summarization, and entity extraction, are currently available for free, eliminating the need for engineers to build these functionalities from scratch.*
01:22 📚 *Retrieval Augmented Generation (RAG) is highlighted as a common AI use case, facilitating large language models to provide answers with additional context and knowledge, crucial for establishing knowledge bases.*
04:39 📜 *Gradient's document summarization feature simplifies the process of generating summaries from provided text, with options to adjust the length and train with examples for improved summarizations.*
08:36 🔍 *Gradient's accelerator blocks, such as sentiment analysis, personalization, and entity extraction, can be combined to build more complex AI workflows, showcasing flexibility in workflow creation.*
Made with HARPA AI
I actually don’t mind the “shocking” titles - adds excitement to the AI space. Common sense and logic help identify and determine the premise of your videos, which are absolutely top among what’s out there. People, say thank you and move along - and be grateful for all of the great content! Geez
With all respect I have a different point of view on that… I do mind it since it is NOT logical and common sense to add “This SHOCKED The ENTIRE Industry!!” In the video titles… it is clickbait and false information… unless he actually shows proof of the whole industry on a groundbreaking research… and those silly reaction faces on his thumbnails is the top of the cake… but I get it why he does this… to gain more views.. I just want honest videos without those attention grabber contents
Is it Local or is it a cloud product?
You can run Gradient in your preferred environment depending on what you're doing. For example if you're enterprise and you want to run it on-prem you can or in your VPC.
Love this! Great stuff Matt!
I would suggest maybe redoing this video with a public domain book rather than something in copyright - wouldn’t want to see you being a target of the Rowling lawyers!
Hey peeps. Ollama for Windows is now available :)
Matt, your videos are excellent, but every RAG test I've seen you do has been a serious softball. I'd really like to see you do a more serious test and demonstration of RAG. Your MemGPT tests included, and MemGPT would be a good place to start.
Thanks for the feedback
great content! thanks
The Harry Potter summary is very wrong... in first chapter of frist book Harry is not even aware he is a wizard...
Unsubscribed. I felt like I found a gem of a channel before, but these underwhelming, forced hype, clickbait paid ads are not what I signed up for. When I subbed, your content was useful, concise, and delivered what the title promised.
You’re venturing into tabloid-type waters and I no longer feel the same quality, honesty and usefulness. Hopefully that changes one day
Thanks for subscribing when you were. Feel free to check out 95% of my unsponsored content.
Hey, at least you still have comments and not using polls as a proxy means of communicating with your audience, unlike some other AI related channels that will go nameless
Just a word of encouragement amongst all the negative comments. Yeah, this video isn’t great due to the lack lustre product but as a UA-camr Matthew is still finding his feet. I watch this channel because I can relate to the guy making the content - I can also relate to the need to take on sponsors and I can definitely relate to the process/struggle of figuring out how to run a small business (or channel). Me - not unsubscribing 😊
Thank you
This comments means a lot amongst all the negative comments. I’m trying my best.
@@matthew_berman I’ve been a web developer since the mid nineties and have considered sharing my experience via a UA-cam channel - however, like you, I’m one of those “neurodivergent” people (very focussed - a little bit ‘different’). I don’t think I would personally do too well with the comments sections - so, I remain unknown. Much respect to you for having the courage to keep sharing via this channel and creating new content (which is really hard).
Click bait works in the short term but will hurt in the long run. Click bait definitely increase click rates but the quality of the viewership might go down and some long term supporters might just leave the channel. IMO, a channel with great content does not need click bait at all and will be better off long term without it. Let your loyal followers know exactly what they are getting before they click on the video.
Miss the old Matthew that focused on open source stuff.
hahaha, dude needs to eat, you gona pay his bills?
I’m still here
Give him some credit, what open source stuff hasn't he covered? Why not recommend him something instead of judging from your porcelain throne
Upload my documents. Uhhhh... No thanks.
That's actually good🎉
If this continues, I stop watching your content Matthew. Please make sure, you make useful content and point it out without clickbait. What is shocking about this?
Can you please explain ?
I agree, all the titles with clickbait, is awful.
Agree. Not great. This ‘product’ is so basic anybody could create it with something like Bubble. Disappointing that Matthew is chasing the $ promoting stuff like this.
Changed the title
@@matthew_bermansimple way to build simple llm apps.
Phenomenal
Lmfao all the AI UA-camrs are putting “shocking” in their titles just to piss off their audience I love it 🤣🤣🤣🤣
Every UA-camr right now: 'SHOCKING new AI tool literally breathes for you! 😱 Not clickbait! #AIOverdramatics
Nothing is free...
Nothing shocks us
I'm shocked that you said that
Matthew, please read the comments and stop using these horrendous clickbait titles.
Got it
Yeah im gonna agree with thr others. Its just copy and paste at this point. Nothing special here, same old click bait.
There's just absolutely no reason to use these "accelerator blocks" ... It's really silly.
The whole AI enthusiast field is SHOCKED by this viral spread of sh*t titles designed to game the UA-cam algorithm. If this formula is gaining you followers then crack on - it's hilarious! 🤣
Firrrrrrrrrrrst!😂
Another AD video
15% dislike ratio on the video. Well I'm not watching this either. Matt usually does great content, but nowadays I only see sensationalist crap.
Getting sick of the clickbait
Thanks for feedback
@@matthew_berman I apologize for the tone of my response, and I hope the irony isn't lost on you of getting upset at the comments by saying "this can’t be your first time on UA-cam, right?"
But yes, I could have spent more time, like I am now, doing more than telling you why I unsubscribed from your channel. The interaction does seem to encourage people to just unsubscribe without saying why.
I'll try to remember to just do that and not provide the feedback (that you thanked me for?) in future. Or maybe I'll ask an AI agent to post politely on my behalf.
Have a lovely day, and I'm sorry Moloch pushing clickbait led to our reaction and to yours in turn.
Please stop watering down your own content with everything being "shocking" like every other AI tuber. It's cringey lol
HI, it is nice, but would like to see (or learn how to build) an onprem alternative.