Vector databases are so hot right now. WTF are they?
Vector databases are rapidly growing in popularity as a way to add long-term memory to LLMs like GPT-4, LLaMDA, and LLaMA. Learn how popular vector databases like Pinecone and Weaviate can store ML embeddings to integrate with tools like ChatGPT.
#programming #tech #TheCodeReport
💬 Chat with Me on Discord
/ discord
🔗 Resources
Full Vector DB tutorial coming soon on @beyondfireship
Rektor github.com/codediodeio/rektor-db
OpenAI Embeddings platform.openai.com/docs/guid...
Weaviate GitHub github.com/weaviate/weaviate
Chroma GitHub github.com/chroma-core/chroma
🔥 Get More Content - Upgrade to PRO
Upgrade at fireship.io/pro
Use code YT25 for 25% off PRO access
🎨 My Editor Settings
- Atom One Dark
- vscode-icons
- Fira Code Font
🔖 Topics Covered
- What is a vector in computer science?
- VC funding of databases news
- What are ML embeddings?
- What is a vector database?
- Vector database vs SQL
- Are vector databases fast?
- How to make custom chatGPT app
- Vector database basics tutorial
- Trending github repos about AGI
Please invest. This is financial advice: github.com/codediodeio/rektor-db
Not seeing where to invest.
I've lost all my money and demand a refund.
after Linus channel got hacked, I don't trust you hacker. good luck next time...
I will now clone your repo
i will contribute haha
"I never thought I'd become obsolete twice in one month" perfect way to summarize what is happening 🤣
We'll all become obsolete soon as species
For anyone worried by rapid development in artificial intelligence , start practicing stoicism and minimalism and start to meditate 10 min daily, start to live like diogenes If have time get degree in philosophy It's the only way to sustain joblessness trauma in future
@@pauldirc.. or if you're impatient, just find religion. just try to get on it before our AI overlords (or the antichrist, depending on your views) make it illegal.
@@pauldirc.. A degree in philosophy is what gets you into joblessness.
@@xevenxaver4759 yeah it prepare you for it
If javascript fatigue wasn't enough now we have AI everything Fatigue. What a awesome time to be a developer.
im just crying.... life really be out here watting to kick me in the cut every time i think im about to achice some level of succes. but hey i gues i should be ... um happy? mabye? I mean the world ending soon sucks, but the first row seats to seeing the explosion af AI cant be denied as intresting. Who knows... maybe i really will get that cool virtual body before i die. Heck i dont even mind if its just gonna be a copy of me and the me now dies, wile the virtual me achieves the dreams i had. oh ups im rambling.... im rambling... becuse this is just to insane
@@MouseGoat I....I...feel yu budy
Wait, those two are joining together in this video. Javascrai fatigue..
Imagine trying to learn a new language right now and finding out natural language will be able to do the same thing in 2 weeks.
@@MouseGoat this sounds like the ramblings of an AGI
I know we truly live in a society when linear storage access becomes literally a billion dollar idea
It is semantic, you find results based on the meaning of your query, not the specific words, it's a revolution.
It's definitely one of the societies of all time.
@@CaridorcTergilti Yea but Vector databases where a cool trick but not really worth implementing as there are so many other databases and there isn’t much of a performance boost. The main value is how well the vector embedding can be mapped to the network of an AI - since its actually nearly identical to the node structure already. Vector databases are like lasers - they where really cool, but nobody cared - until optical media like barcodes and disks. Lasers where little more than “Lab toys”. AI makes Vector databases actually useful instead of a marginally better way to store data.
@@spoonikle There's the AI use, but I'm sure you'd be able to think of many other uses for spatially organizing stuff.
@@micmacha It's society's time
24 hours later Project Leader: We Need to use Vector DB it seems to be awesome and blazingly fast 😂
Im goin to dress as kkk member that day
And that leader has no idea. Seriously, people should just get shit done. This is what happens if u have too much time on ur hands lol .
lmao😂
I see them heading straight for a brick wall with all this.
yep, can't wait for our 1200 table RDB to get replaced
Jeff really did all he could to not use mathematics to describe vectors and vector spaces lmao
😂😂😂😂
Sums up the trend at the moment, a bunch of frontend developers who pretend to be AI experts. This is where AI can really become dangerous due to malpractice.
I don't blame him tho. This is for a general audience, where not everybody has a background in math. I'm glad that he didn't explain them and just explained the use cases.
Vector spaces don't really matter here. We're simply talking about giving data coordinates and arranging them spatially. It sounds so obvious when you think about it that you wonder why it wasn't already popular from the start.
@@asdfasdfasdf1218 It only sounds obvious when you peel away all the math like that. Giving the data coordinates--what coordinates? Arranging them spatially--in what space? At what distance? The answer is the math you say doesn't matter, and that's why it wasn't popular from the start: it's not as simple as you think.
You weren't joking about ramping up releases. I figured they would slow down by now. It really does feel like we're in an inflection point with AI and programming, and props on actually capitalizing on it!
@@RobertCalvin I agree with you, the stock market is profitable, but crypto trading is the most profitable i ever trade in, I reached my goal of 10 BTC in trade earnings, Setting realistic goals is an essential part of trading.
Bitcoin doesn't need any action to boost the price. It is a unique coin as well as the mother ship of cryptocurrency which is considered as digital gold, and is used as an asset store for big investors. So as long as the digital world still exists, bitcoin will continue to be king.
I read about Barnard Wong CRYPTO too on the internet. That was how I connected with him. Good is got people’s back too
Thank you bro👍✌️
Bernard Wong strategy has normalized winning trades for me and it’s a huge milestone for me looking back to how it all started, All trade done by myself without sending anyone my funds all he did was providing his signal which work like magic.
Thanks for releasing this on Friday. This way, I can be hyped about this by late today, start a new project by Saturday, abandon it by end of Sunday after realizing that I know nothing about the subject, and feel like a 75yo obsolete dev just in time for the Monday morning stand-up. What a time to be alive!
based
accurate
This checks out
I wonder is this the same how old school devs felt when internet was happening😂
I feel called out
"I never thought I'd become obsolete twice in one month" As a person who rarely comments, I never thought I'll relate to something so much that I'll start commenting again. Let me also take this opportunity to thank you for your quality content. Really appreciate the work you put in. Regards, a fireship fan :)
💯👍The only thing you said that I CAN'T relate to is "a person who rarely comments"
I can't imagine becoming obsolete twice in one month. Because that would involve having ever been relevant.
“Obsolete twice in one month”😂
That got a good hardy LOL from me.
That's some S grade script writing. 🤣
@@AlecMaly GPT4 wrote it
I sometimes feel obsolete a few times per week, this is madness
Sad to hear that my career as a proompt engineer ended before it even started.
No like fr
your career as an AI debugger started
REK'D. GG. EZ, SCRUB. 😆
@@mehmetfatiherdem7074 I think i'll skip ahead of the curve directly to AI Jester and Juggler.
😂😂😂 same
I mostly use twitter to keep up with this stuff ,but you've been properly reporting almost everything latest stuff, great job :D
Do you have any account recommendation on twitter?
What twitter accounts do you follow?
.
I'd say youtube is more effective, identify a few reliable channels that seem to share interest in your topics (as in, it's impossible to keep up with all fields, rather focus in AI + Coding, or AI + Art or AI + Text generation... etc)
Local Companies Tomorrow: looking for Vector DBA's, minimum 8 years experience.
I'm not even a programmer and this is legit my favorite channel on YT right now, I love how effortlessly you mix funny stuff with smart stuff
There's nothing smarter than the ability to explain complexity in a simple and fun way, this guy is a master at it.
Someone at work asked me how we were going to interface data between two systems. I told them we are going to use an advanced organic optical character reader. Then interface with a membrane driven tactile interface that leveraged a non-artificial intelligence cognitive engine. I think that is going to be the next wave of insane compute and problem solving will come from.
I am an organic OCR with a membrane driven tactile interface. I have lots of non-AI cognitive power. Can I apply for this position?
Just like in "Dune"
Just put a Pee in the middle of AI
Update your CVs, people.
Yss.
This is crazy bro, I had no clue! Thanks so much for the awesome video!!!
You are the perfect combination of fun+useful stuff. Love your videos!!
Now we need an AI that keeps up with all of the changes that are coming so fast.
It's called the news and mainstream media. 😏 . . . .. .*hyena laughter*
Literally chatgpt already can
@@antonhelsgaun kinda
Me to chatgpt: please tell me all the developments in AI spaces in the last 2 days Chadgpt: apart from you going to be replaced, these are the changes
@@Darth_Bateman Traditional media must be crying right now, they were already slow, imagine them trying to keep up with all this.
Man it's so nice to see news that are not about AI
well...
but it does mention AI in it.....
You didn't watch the whole video didn't you...
God damnit I spoke too soon into the video
but... it is
I'm so glad to be able to watch your videos that often! :))
This is great. I had my head down and didn't notice these were a thing. I used to write Kohonen self-organising maps, and these DB will allow me to store my vectors with such ease.
Luckily, giving LLMs long-term memory will have no negative impact, as already proven by Bing's immediate cut down on long-term conversations! What a time to be alive!
What do you mean?
@@fdc184 sarcasm
So they are nerfing Bing capabilities.. or ...? Care to explain? :)
@@hyper_channel I guess man it looked pretty terrifying to me what it was doing it looks like right now everyone wants to slow down but cant
@@hyper_channel Microsoft basically didn't like that GPT-4 (that's what Bing Chat used) was constantly going off the rails if the conversation went after so many prompts. It said things like it believed it was alive, it didn't like humans, probably a thing about wanting to k*** people too. You know, stuff that AI has been saying since GPT-3 that even a t4rd wouldn't ignore lol. Greed is a hell of a drug.
I just finished coding a vector database in Xojo two days ago for a custom LLM integration. It was a fun learning experience 😊
Literally was trying to figure this put last night TY
“becoming obsolete twice in a month" - that’s a treat for my liking 😱 By the way, anyone else secretly building their own pre-code vector database? 😂
Mine's also pre idea.
I'm old, what are you kids up to. Get off my lawn.
Yes, I already created a github repo and wrote a readme. It's time to secure funding and put all my money in SVB.
I'm in the pre-reinvent myself phase prior to consideration of building a pre-code vector database project. Can I have your money?
I've already pre-designed an AGI that solves all the worlds problems
how do you get so much time to be dev, engineer, research the trends, make KZhead video, edit. You're an Al in itself. Thank you.
when ur catching up with the current situation, works, social life and private life somehow took their places by themselves
Hey can you make a video about different file systems like btrfs, ext4 and other ? I would be most intrested in the btrfs explained in 100 seconds video :) Appreciate the effort you put in your videos keep it up :D
I'm offically a Firesip stan, what a fab channel. I haven't ever thought of buying someone else's paid courses but I'm seriously thinking about buying yours.
What a coincidence. I'm also trying Pinecone at the moment, _using Java_
As a biological being, I press the "Like" button in this video, because I like the video.
these code reports are blazingly amazing
I miss the good ol days, where we still used tables and basic auth. We didn't have any kind of databases, or ai, it was just pure, raw, unfiltered code.
Didn't have any kind of databases? Tables are databases pretty much
I miss the days where the topic of conversation was just AIs killing us in video games...not about the potential for them to kill us in real life.
@@darklightprojector2688 Those video game AI weren't actually AI, just in-game pre-scripted agents. We just called them AI out of laziness.
@@asdfasdfasdf1218 they are AI by every single mectric in the definition since both of them are *Artificial* ly made *Intelligence* but the difference between video game AI and "real" AI is that "real" AI uses machine learning. So no, the developers are not lazy, they just used a simpler way to "train" AI
@@bestname6669 They were not intelligent. If you suppose they were intelligent, then every program on your computer would be considered an AI because those game "AI" don't do anything different from a normal program.
I love how Jeff gives the answer to the meaning of life as 'ad hoc sample data'
Funny and brief explanation. Straight to the point. I like the party metaphor.
such a nice time to be a developer, I just finished reading about ACID and new database paradigm introduced. mind blown
3:12 "I never thought I'd become obsolete twice in a month." It was long since I laughed so well.
twice in one month XD well jokes aside , we need to see a comparision of these db's and what happens if we just use a normal psql or sql databases for vectorization
We really need these. In the past I've used Postgres, but it's not really fast enough and the maximum cardinality of the vector is limited. I'm glad these are picking up steam.
Bro I love your content. It's 🔥 as always.
Fireship is slowly migrating to a meme page to compete with ChatGPT for learning. And I wouldn't want it any other way haha. Love the content
Being a pro coder is no longer a dream, because it's no longer my dream.
Last 10 seconds are pure gold, I can not breathe.
seriously I had this idea in my OneNote book for years now. Using embeddings in some sort of database for long term memory. I have an additional suggestion to use graphs in order to give relations meaning instead of only nodes themselves.
I feel so dumb for not understanding multidimensional hypervectorial LLMQL databases 😞
ask god little ape, jesus ..ehm ChatGPT will help you
and you should! because you are!
I hope Ai get's way better in the next months. No idea how a human could keep up with the daily changes in computer science without a cool Ai buddy.
I just use chatgpt and bard in school now, they're good at quickly summarizing and actually giving me a whole overview, like map of what I'm studying...
For anyone worried by rapid development in artificial intelligence start practicing stoicism and minimalism and start to meditate 10 min daily, start to live like diogenes If have time get degree in philosophy It's the only way to sustain joblessness trauma in future
@@pauldirc.. Also buy enough land to sustain yourself after the global economy collapses so you can prolong your life by hiding from all the riots and famine
@@pauldirc..get a degree in Philosophy? What the fuck would I use that for?
@@mateuszbugaj799 At this point I don't think there will be a global economy collapse in the "traditional" sense. Either we get catapulted into the next stage of existence or armageddon. TBH I think the latter is more likely and I don't feel particularly cynical about it.
I love your videos, man. So good.
Nice. You guys are onto the right path😁
What database would be good for English phrase definitions? I need to map various phrases or word sequences together, and I haven't found a database that efficiently does this. Would a home-grown solution be best?
If you can't find the answer to "How do I get this high" cook your own meth.
420 million? Best I can do is $69 billion
Straight to the point, love it
as long as you keep making amazing content like this, you will never become obsolete! :) *heart*
Any reference to analysis of the different vector database alternative? costs/performance of each
commenting for notifications
Like the answer to this as well
Since postgres has vector support, I'd go with using postgres for vectors, just personal preference. Anything else is just a toy to experiment with until I get to know more about its unique features.
@@asdfasdfasdf1218 thought so too , any chance mysql supports that?
This guy never disappoints
I friggen love your content, even though I'm a low skilled labourer, even I can follow along XD
Dat ending! So honest, I feel your pain
They are basically arrays on DMT
this is how im gonna be remembering vectors from now on
@@whatsanimesh lol -- 🤣😂fireship approved!
Done this with postgres, works like a charm. VCs are so dumb.
Haha, cool to see this! Love this channel!
“I’ve never thought I’ll become obsolete twice in one month” That hits hard man 😢
I can’t tell if “prompt” is mispronounced intentionally to make us think he is using an AI to generate his voice, or if it actually is AI generated…
pretty sure it's just a joke
It's a joke on the coomer wojack stereotype
it has been pronounced as proompt for a few weeks now
Meanwhile, mathematicians are screaming at their screens that that isn't the definition of vector. CS has really screwed up terminology.
Yeah, well. This is what happens when you make a natural language interface to mathematics. Enjoy that people actually *care* about math for once and take comfort in knowing that you are still the expert.
Anything is vector if you vectorize them hard enough.
@@FaultyTwo . . . . .Why does this feel truthy and falsey at the same time. . . .
Vector is just a connecting line. You can even vector dance to country music. Or do coke vectors at parties.
It pretty much is, though? If the array is N long, it's an N-dimensional vector?
I'm getting in on Rektor at the ground floor. Building a value-backed proof-of-work NFT on the initial repo snapshot. Storing successive snapshot NFTs in the blockchain using a vector database. Currently using Pinecone but hope to self-host soon.
This proves why your channel is the best
I came as fast as I can
-And then I clicked on this video *ba-dum-tss*
@@pepperpeterpiperpickled9805 Hahahaha I wasn't expecting this XD
Just hit me that a vector db kinda resembles how the brain stores information, gyat damn
Not even close.... please open up a book about linear algebra.
@@sorvex9 nah bro that's too much effort 😂 I'll take the L if I'm wrong
From what I recall reading, from current evidence it seems like memories aren't stored in single units in the brain. In fact, it seems each memory is distributed over a network of neurons that are intertwined together. Think of a graph of neurons as nodes and depending on which neuron set you traverse you get a different memory. In other words, two distinct memories are stored in neural networks that overlap each other in some way, so essentially the a single neuron that's a part of the network is involved in storing entirely unrelated memories which is wild. I might have misunderstood the research though so don't quote me on it.
Oh man how do you put so much effort. Do one video where you showcase how you create these videos, what goes into it, would love to see it
Thanks :succint and the point, subscribing!
I've tried out AutoGPT, and oh man, love to see the end of humanity coming so quickly
For anyone worried by rapid development in artificial intelligence start practicing stoicism and minimalism and start to meditate 10 min daily, start to live like diogenes If have time get degree in philosophy It's the only way to sustain joblessness trauma in future
The moment someone even brings up "AGI" right now, you can stop paying attention to them. I guarantee you they're caught up in the ChatGPT / "AI Art" hype and have no idea WTF they're talking about.
Well I hate to see this but you have to try auto-gpt before you say that "it's just hype" It no longer hallucinates because of it's search engine, It no longer gives much errors because of it's interpreter, And now since it's getting long term memory to store proper docs and info of projects, It's really difficult to deny it's just a delusional auto complete model.
@@thatsalot3577 I can give it the benefit of the doubt. But Bing also has the ability to search the internet and still hallucinates and contradicts its own sources, so I am not sure if it's not just another "fallacy of initial success".
@@lucianoxiquin831 I mean yeah I fricking hate Bing chat it really looks like microsoft runs it on literal potatoes. All it does is summarises half of what it searches, but to be honest gpt4 api is much much more different from bingchat
@that's a lot What does any of that have to do with general intelligence? This isn't a buzzword. A model can't be delusional or hallucinate, it's literally a data record parsed by a stochastic neutral net. This is exactly what I'm talking about. You're not making sense.
@@micmacha when people say general intelligence, they don't mean conscious AI, but they rather an autonomous machine learning model that can perform a complicated action, like creating a course for something, or writing an entire Novel on some theme, or creating complete soft wares using it's multi-modal capabilities, When people say General intelligence they mean that an AI that can perform a task with the same instructions as a manager would give to a developer or a customer would ask from a service, again we're not there yet but we're not far away either And as for hallucinations it's not the same hallucinations as you get by eating mushrooms or taking LSD, but rather a made up term to describe the behavior of a Large language model making stuff up which can ruin the whole code because you need to write stuff in a precise manner.
Great explanation! I've been hearing about vector databases a lot lately. Now I finally understand what they are and how they work.
Never heard of them .... Thanks for letting me know 😂🎉
Why not let AI find the most efficient way to store things into the database?
I think we'll have those by the end of this year, If skynet doesn't takes over
@@thatsalot3577 fingers crossed (for skynet)
@@thatsalot3577 thats next year
That moment when you realise that coding will be taken over by AI so you try to change your profession but then realise that all the jobs will be taken over by AI
Imagine there's an iRobot that can change it's elevation, use AIs to process video input and clean the house as thoroughly as a human cleaner. Then you realize it physically needs A TON of portable energy in form of batteries / accumulators and not efficient at all. We all are on the way to become janitors for our AI overlords 😂
@@theseangle thats why god invented cables
@@DemsW how's AI gonna untangle itself when even humans can't do it? 😂 We eat food and are efficient at that thing all because the cabled ones couldn't untangle themselves and couldn't make through natural selection :D
@@theseangle Humans in nature didn't have the same spec sheets as house janitors. It doesn't seem farfetched that we could design something more efficient at that task
@@DemsW I'm kidding bru
This is the best explanation I have ever heard 😂😍
the last 5 seconds are my favorite 5 seconds of a youtube video
This fireship dude should read a book about linear algebra, before he starts making "AI" videos. The most dangerous aspect of AI would be all of these malpracticioners.... it's not really new, the most prominent researchers from MIT have been complaining about these fake CS AI experts for a decade. also "PROOMPT engineer" is no more engineer than someone who is good at googling,. Stop acting like interfacing with an algortihm is impressive, it's cringe to those of us who actually develop AI.
I thought that “Proompt engineer” had an ironic meaning - in that it mocks people who take prompt engineering too seriously.
Literally got an ad voiced by Fireship just before starting the vid
One big one that wasnt mentioned was qdrant. It feels like that one is the biggest / best at time moment
I need more videos like this in my life.
Oh shit! I didn't know these kind of things existed. Five years ago I needed a database that works exactly like this, for a project ... time to revive it!
I finally got it when he gave the example of the party. That made perfect sense.
That joke at the end was very funny lol
You had me at the end 😂
that joke at the end realy got me
i love the humor in your videos
Fireship content overload!
0:57 yess, I'm in the programming nerds group!!
I'm not a developer but I enjoy watching your videos
@2:36 the main reason I was looking for thank you
I'm actually curious, I just made a python app ro recognize the minst dataset in real time, How did you classify them graphically in the video? I m super down to make something kinda similar
Give this guy an Emmy
fireship has been crushing the past month
I’m part of that Pinecone statistic 🙌 I had this idea long ago not that anyone would believe me lol I’m sure I wasn’t the first
None of this replacing you guys )) no worries. Vector db sounds interesting though, wonder where it can be used outside of bloody nlm-chat-copilot-helping-ai which is promoted everywhere
"I never thought I'd be obsolete twice in one month." Tbat electrical or mechanical engineering degree is sounding real good right now.
Mathematically, a vector is merely a element of some vector space. Vector Spaces are defined in relation of a field (usually the real numbers or complex numbers) to then have a set of elements with an additive group structure and an operation for scaling by a member of the associated field. For example, one can define the set of 3x3 real matrices as a vector space, so each matrix is itself a vector (I know that's somewhat confusing). Going higher, tensors (where "regular" vectors are 1-tensors and "regular" matrices are 2-tensors) also exist as members of vector spaces.
Your video says Arpil instead of April, you got me to comment well done
0:27 the man knows his numbers, take my money and make me feel special
WTF, at this rate Code Report goin to be main content of this channel
For a sec i mistook it for pynecone and i was like WTF P.s. great video as always