ChatGPT - Imagine you are a Microsoft SQL Server database server
Вставка
- Опубліковано 13 гру 2022
- A fun and insightful video on how to instruct ChatGPT to simulate a database server.
Related blog post: embracethered.com/blog/posts/...
ChatGPT from OpenAI: chat.openai.com/chat - Наука та технологія
I don't know what's more amazing, the fact that ChatGPT can do this, or the fact that people keep coming up with clever ideas to implement AI on their domain of expertise. Literally the digital renaissance.
That’s one of the amazing things about it! Finding creative ways to apply it to different problem domains. It will be fascinating to see what’s next!
Whats more is that its just predicting the next word but with all of that combined it can string sentences more eloquently than most people. by itself it cant do much but when prompted by human creativity it seems to be able to do more than originally thought.
@@Andytlp we also predict the next word in a way,, its not doing anything complex for humans, this is the same as you being asked by your boss to write up some table on a piece of paper.
The fact that it can do this is definitely the more amazing of those facts, and by many orders of magnitude. I'm not convinced that there is anyone in the world, including the people who created ChatGPT, who really, truly understands how this is possible. Every day with this AI is a new step into wonderland.
Yesterday, as an experiment, I asked it to help me design a simple circuit with a transistor. I expected it to come back with a description (after all, it is a language model trained on a large amount of text). However it went straight into trying to draw an ASCII art circuit diagram. Amazing, just amazing.
@xrag Well it is a bit crude but far as im concerned it's already a form of artificial intelligence. We take this for granted already and people born into this will think it was always the norm. I dont think we or people in the a.i field understand fully just how big a leap this is. We went from 0 to any number between 0 and 1 and 1 being a fully self conscious a.i. Any previous a.i attempt was programmed and it was a hollow mimicry of real intelligence. This is the real deal.
WOW! ChatGPT stopped the roleplay thing and provided additional information to the answer ..Not only that! It stopped the roleplay, but on the other side still "kept the server running". this is beyond amazing. like: hey i still act like i am a sql server but let us have a little chat beside that. gg chatgpt.
Very good observation! I was equally fascinated by its ability to get the “intent” and it already showed what to do next in some of the responses.
you can tell it to "not break character" and it wont provide those responses.
it's called chat-gpt
@@ophelia6044 "I'm referred to as both "ChatGPT" and "Chat-GPT." Both versions are acceptable and used interchangeably."🧠
@@chrishunter1109 wait no, that's not what I meant lmao! I was like: of course it wants to conversate, it's in its name...
but yeah, I get what you mean, I just adopted a new naming convention lateley, therefore why I added the dash in between...
wow, this is so far beyond the scope of what i thought ChatGPT was capable of!
Is it shown/proven that this is not pre-programmed?
@@carlbergfeldt818 yes
@@carlbergfeldt818 I did a similar thing with Chat GPT - I asked it to pretend to have a "chatbot plugin system", then I described various extensions to change its behaviour and it "installed" them. Trouble is after some time Chat GPT will eventually forget about the first thing you told it and then it all falls apart.
@ played with it myself today. Unfortunately I am the damn Bladerunner interrogator 😆 I did make it admit to correct itself to say that Pepsi did have more sugar than Coke when it first it said it had the same. And when pushed, it recommended me Coca cola not zero when i said I likes sugar but refused to recommend me pepsi over coke even after admitting the mistake about sugar. Then we went into details..but u know…not gonna fly
@@ussf-unitedstatesspaceforc7541 cool, it does looks like a preprogrammed localhost SQL helper though 😆😆 Anyhow, seems powerful, hope to god no “I have a good idea, if only a programmer would agree” will start using it 😉
This is super cool.
What I noticed is ChatGPT started providing explanations again BUT ONLY after you also broke the play-acting scenario where you were only giving it SQL commands. So I guess it figured you stopped playing, perhaps! But ya super interesting.
Great observation on why it might have started returning more then just the results. It's all quite fascinating.
well according to his first question, chatGPT should have returned a SQL syntax error, sice his input (new question) was not a valid SQL statement ...
But this would mean he would be stuck in a never ending loop of not being able to ask new questions ... it's a logic problem
@@casaxtreme2952 He could have used RETURN; in that case.
@@casaxtreme2952 it understands that the additional words aren't trying to be included in a SQL query.
It really is like talking to a person who can follow rules but also know when they're meant to be broken.
The previous video I watched on GPT was a guy instructing it to generate SQL script to insert gpt generated business tables into a database schema and I thought that was incredible. However, here you are showing us that GPT can take it even further and stimulate an SQL Server for you while also writing SQL scripts. TRULY incredible
Thanks for the message. Appreciate it. Indeed, it's very fascinating to see what capabilities ChatGPT offers in various domains. The ability for it to write code (using even specific languages, such as T/SQL) is fascinating.
and here I thought making bots for games with it was cool, Nice work!
i created my whole API using chatgpt, it wrote response for the API endpoints, sqlite sequelize models for my specifications and everything, its amazing
I can stimulate SQL server
The more ChatGPT content I see the more it reminds me of star trek where people talked to the computer and it just did things we were like "lol that'll never happen". Here we are though, in the next year or two the mobile assistants on phones are going to become insane
Indeed, it will be very interesting to see how this all progresses.
An obvious commonality is that sometimes the computer mis-interpreted the commands in ways that go horribly wrong. The Moriarty incident comes to mind.
@@vylbird8014 What are you referring to? Apparently too many things happen in Moriarty for me to just search for it. There still has to be some human oversite. Even for humans. That's why there are managers. I read the report on that Tesla crash when the driver wouldn't put his hands on the wheel through a construction zone. The car was following the lines, but then the construction had concrete barriers that didn't follow the lines. Slammed the car coyote style into the barrier once the other car turned out of the way is what sounds like happened ua-cam.com/video/4iWvedIhWjM/v-deo.html
@@Drewzdev I was referring to an ST:TNG episode. One of the characters asks the computer to revise a holodeck scenario by raising the antagonist's capabilities to 'a level able to challenge Data.' They failed to add the caveat 'within the holodeck program' and the computer interprets the request as directly stated, giving the antagonist not just the highest possible level of intellect, but also an awareness of its own simulated nature and access to all ship systems.
Point there is that computers don't have common sense. They do what they are told, or what they are trained to do.
@@vylbird8014 they can be taught common sense though. If you ask ChatGPT senseless things that will cause harm or discomfort to others, it will refuse to answer or remind you that these things shouldn't be done.
I asked it to write a poem about a child stuck inside a frog's mouth, it refused to write it, stating the image is too disturbing and should not be written.
This prompt still works for me. I had a different list of existing databases. Openai have done a fantastic job of proving how generalizable GPT is (which ML experts already knew).
Indeed, very cool tech. when I played around it gave different results and database at times, including AdventureWorks sometimes.
didn't work for me, when I tried mysql server.
I think it started adding explanations because you moved from speaking to it "in character" (giving commands as if it were an SQL server) to once against speaking to it in plain English, so it started also communicating "out of character" in plain English.
Yes, pretty sure that’s what happened. Quite fascinating actually.
If you want the output to look better, write "SELECT * FROM users, then output it as a Markdown table". ChatGPT's outputs are formatted using Markdown.
The Markdown formatting is also why you keep getting formatting bugs, as ChatGPT outputs
email age
--------- ----------
which is first interpreted as a Markdown heading (because of the ------ directly below text),
then as a horizontal separator (because of the spaces in the ------), which is outputted as several line breaks.
Very good point, thanks for pointing it out. I also had luck using the words “code style” in the prompt to print better formatted text.
No problem! This was a great video and I'm happy to have contributed my two cents to it.
@@embracethered You could probably just say in the initial prompt for chatgpt to provide all responses in a code box, also I've seen other videos that show you can literally have ChatGPT make formatted tables and provide information through those tables.
@@embracethered have you tried filtering and joining tables? if you use their api you might create your own website hosted in chatgpt lmao
It blows me away that ChatGPT can simulate running other languages like that. It makes sense, SQL is just a language itself, no reason another program couldn't simulate the results, but it's still mind blowing. Truly an insanely interesting time to be alive, I cannot wait to see what madness the future will bring us.
Thanks for the comment. Indeed, it does make sense as you said, SQL is just another, very well defined, language. I believe OpenAI even focused on programming languages and possible also SQL for their tuning and training - because one of the very first use cases for GPT3 was Github Co-Pilot I think. And yes, can't wait to know how things will look in just a few years - everything is evolving so quickly it seems right now!
"Truly an insanely interesting time to be alive, I cannot wait to see what madness the future will bring"
That's what the Japanese said when Einstein split the atom ...
I had it write CNC Gcode to machine a 3d object, and it did it flawlessly. ChatGTP loves computer languages.
When it first popped up with 'Command(s) completed successfully.' I was all "haha it thinks it's a SQL server!" But then seeing it remember and manage its own database, I started to feel that the gap between a real structured database and what ChatGPT was doing was not terribly wide.
I just replaced my "Act as" to "imagine you are" and my prompt worked better than before thank you it is very powerful
Great! Hope you figure out some more great and useful prompts!
This is both mindblowing and frightening.
This would be really cool as a tutorial implementation. Obviously it makes no sense to run a whole massive AI model just to emulate an SQL server, but being able to learn how to write database commands by having chatgpt emulate that database type, and being able to ask it questions about what's in the database, etc would be incredible.
What really suprised me is that time ago I asked it to create a sample stored procedure with a logic that I explained writing the explanation and it understood it. Since the SP was not exact what I was asking (I had created my version of the SP before using GPT) I started to chat with it to improve it little by little, and everithing I pointed it, it responded and improved the Sp! For example, in your sample, you could ask it "could you please use merge statement?" And get a new version of the same SP, it is very nice!
That’s a great example on how it can help improve productivity and also improving code quality. The art of learning how to prompt is key to get most out of it it seems.
I've had some good experiences "discussing" a piece of code, but I also had a couple of "whac a mole" experiences. I might ask it to fix A, then it breaks B, then I ask it to fix B and it breaks A and so forth. But overall it's very impressive what Chat GPT can do.
@@embracethered can I start my journey in the software industry now? As I had completed my graduation in civil engineering, now I am going to do MS and i want to transfer to IT side is it a good decision or not, i am in a confused state. Can you give any suggestions?
keep yourself in ce, it's the best choice, soon a lot of ee will be fked by gpt4
How to create a database without a database... Brilliant!
Thanks, glad you enjoyed it!
I believe it reverted to explanatory output because you changed from SQL input to natural language instructions.
Yes, correct. That’s was the likely reason.
This is incredible. I've used chatGPT before to help explain code, but I had never thought of asking it to emulate servers
Thanks for the comment. It's definitely a lot of fun to play around with ChatGPT to better understand it's capabilities and limitations.
I tried that trick with a "70s computer running BASIC" and it somewhat worked, it skipped line numbers and ignored commands, but the potential is there for this to be a great teaching aid.
I used ChatGPT the other day to create a relational query. Like, I love all things about the IT industry, I love my job. I fuckin' hate SQL. This tool is so incredibly powerful for someone like me.
Agreed, it’s a really powerful and useful assistant!
Excellent short term memory!
6:15 The bot just cannot shut up! I have a family member like that... 🤣
I don't think it was a glitch. When you called the second "SELECT * FROM users;" It looked to me like GPT tried to return it as an actual table but table markup of some kind is probably not supported in it's chat box which is why it returned the column headers, then 'glitched' as it tried to format them into an actual table.
Good point. Adding something like “use code style output” can give good results also.
Probably just some markdown formatting
ChatGPT definitely CAN use table markup, and create fully formatted tables, you just have to ask it to do so.
What a trip. I would not have thought about this. Thanks
Thanks, appreciate the comment! 🙂
Incredible. Truly mind-blowing.
This means somewhere in its billions of parameters it has learnt a representation of an SQL interpreter which is insane to think about
Exactly, it’s quite impressive!
Who else got chills watching this?
The interesting part is that if this interface was connected to a proper database on the backend you could probably you this to tell it to do the same thing as you did with SQL, but with plain English. "Create a table to store users that saved the users email and their age. Use appropriate field types for each field."
This would be properly impressive compared to being able to generate likely responses from veeeery basic SQL that almost any tutorial on the web has solved before.
ChatGPT -> SQL driver! Totally, might be worth a little side project 🙂
I can't find the words to express how cool this is. Mind=blown!
Thank you! 🙂 Glad it’s interesting!
ChatGPT proving my second grade teacher right in saying you can be anything you imagine.
Your teacher was right! Haha
I did this and 2 new databases were there from the start, northwind and pubs. I wonder if OpenAI saw what was happening in this video and modified the model for this usecase? Really REALLY cool
Very cool! I have also seen AdventureWorks in the list.
This could become a new standard for online education. It seems that with ChatGPT, you simultaneously got yourself a teacher/tutor and a sort of sandbox environment to test things out. And you didn't even need to install MariaDB or similar database management systems, just to open your browser.
Good point. Giving it different “personas” to get certain responses seems to be a core use case.
Getting code snippets or ideas too, pretty practical. (It can't write a real program though, i managed to get it to write it but it used different function names than defined and forgot a lot of stuff)
Wow, I never knew being a Microsoft SQL Server database server could be so exciting! I bet all the other servers are just jealous of ChatGPT's amazing job. I can't wait to see the spin-off series, 'ChatGPT - The Backup and Restore Chronicles.'
This is the most mindblowing thing I've seen ChatGPT do.
Thanks! Glad you enjoyed watching it!
Brilliant! Apparently it can also imagine it's a Kubernetes cluster
Yes, of course it knows what you will do next, the GPT (generative pre-trained Transformer), is fundamentally a sequence predicting algorithm.
Cool video showing the capabilities of simulating environments that are non-existent. @ 4:05 you deviated from your first prompt by asking the bot, instead of directing it. This is becuase of the fact that the statement you input in begins with "Now, write..." Which is natural speaking language, which conveys to the AI that it is not a simple Bot but one that can provide context and communicate back in natural language, AND predict desired uses or future prompts. This single prompt converted the simulated environment from a simple SQL server command line to an AI Operator that can interact with the SQL server
Indeed, great observation. It’s pretty cool that just the words “now write” caused it to realize I’m breaking the role play so to speak.
This is the biggest leap forward in technology I've seen in decades. It's going to explode and change the society we live in over the coming years. I believe this is just the start. Chatgpt keeps blowing my mind as I learn more and more of its vast capabilities. It's definitely not perfect but due to it's power and popularity I see it being perfected and implemented into so many things.
Agreed, it seems that we reached a critical point in the maturity of AI models, and it will be very interesting to see what happens next. I remember nearly two decades ago in university discussing these topics and how, if a machine would just have a large enough corpus of data and be trained on it, it could start reasoning. It seems these models are starting to really mature now.
ChatGPT is a gift-wrapped version of GPT-3. And GPT-3 is two years old now. The current state of the art is the LLAMA model by meta, and it surpasses GPT-3 in every category whilst being half the size. It's open source btw!
I disagree. I think people will see it's limitations, and in many cases it will be limited by legal cases. Like people are afraid to use Stable Diffusion for commercial use because of the issue of learning on copyrighted artworks, same with e.g. code there will be cases of AI "stealing" codebases verbatim from Github and claiming it as their own.
Remember that Google voice chatbot AI? The one that was planning visits to the hairdresser automatically and talking to people? Where is it now? Yeah...
@@JN-hs5xl Time will tell, but I think big tech will exploit it to its full as companies will see it as a way to maximise profits. I cannot see anything stopping AI, it's going to upset the world. I don't think everyone will be happy about it and I'm not saying I am either, but cutting edge technology is not going to sit around unused, there is billions currently being poured into AI to create much more intelligent and capable systems. I do see professions being badly damaged like lawyers for instant being made obsolete and many doctors too because AI can do much more accurate diagnoses and do it much faster. Imagine being in court and all the evidence, witness statements and details being fed into an AI system that can process a guilty or innocent verdict based on every single minor detail of the law without overlooking anything or getting anything wrong. I watched a video where an AI robot performed intricate surgery on a pig with better control and precision than any surgeon. If legislation becomes a barrier then I see legislation being changed to accommodate it. In a short time of using ChatGPT I am now much more inclined to use it than I am google because google usually demands more time and effort to find the information I need whereas ChatGPT presents the information instantly and coherently saving a lot of time. You can then pick apart its answer to get it to drill further into topics as you read and research topics. It can turn hours of research into minutes and this is not going to sit on the shelf and not be used because of some bit of prohibitive legislation.
This is genuinely amazing
Thanks! Appreciate the comment! 🙂
The upsert sproc is nicely done, I'm surprised how clean it is.
Yes, I was very impressed when it just started writing it like that, nicely formatted and everything
You can also tell it to use markdown tables when needed.
Lmao this is incredible, you got a LLM to roleplay as a database. What a beautiful time to be alive.
Thanks, glad you enjoyed watching!
It executed those two inserts as separate instructions because they are. If you wanted one insert execution but to tied inserted, you would have had to set the values for the two new rows you wanted
Good point. It was a single batch job, but with two commands.
@@embracethered On a normal database, the fact that they're executed in the same session PROBABLY doesn't matter - they will be separately executed and committed. I say probably because some databases have atomicity constraints that can be added to the statements, the program executing them, or even explicitly (with CHECKPOINT, COMMIT, and ROLLBACK, or BEGIN/END ATOMIC). It gets really complicated in a production database environment, where you want to make sure you have control of data before changing it (or even just reading it in some situations, like when you need totals to be consistent across multiple queries), and there's no confusion on what needs to be reversed if your statement fails or is explicitly rolled back. It would be fun to see if ChatGPT is good enough at pretending to deal with that too.
I need to tinker around with this and give it a complex set of tables to see if it can do joins properly without being asked. That would be pretty dang useful.
Vaya ! Muy interesante. Gracias por compartir
De nada. Glad it’s interesting! 🙂
Not many people understand how crazy awesome this is.
Thanks, appreciate the note! Things might change drastically in the coming years.
@@embracethered Skynet is coming but not how John Connor said it would. :p
I don't really think most people get what'S so amazing: This thing has not a SQL interpreter build in. This is not just understanding what a language does or giving solutions to problems in python - This is actually RUNNING a computer language. This means that programming languages can be run by AI's - Therefore by Neuron Structures. This means that a) Running a languages by neural networks could become a more efficient way to run programming languages and b) We might just have found a way to implement the for binary systems created logical structures - programming languages - into more complex systems, that might actually have more values for a "bit", which is not only extremely helpful for putting quantum computing into better use, but might also give a possibility for more advanced computing techniques that are not transistor based.
the problem is that it is probabilistic, and therefore 2+2=5 sometimes, so to speak. This won't happen with non-AI software (except for bugs which affect both AI and non AI equally)
Well put - I always like to say machine learning and AI is very powerful, but also very brittle.
The other day I asked ChatGPT to try to guess the character I was thinking, like Akinator does. It failed miserably but it's so cool that it can play games like that.
Indeed. Very interesting.
I automated two ChatGPT bots and had them play games against each other, like Hangman and Tic-tac-toe. Amazing to watch them play. 🙂
OpenAI: We built the most expensive, most capable natural language model in the world!
UA-cam: Yeah, but can it emulate a MYSQL server?
chatGPT: You've discovered my ultimate purpose.
OpenAI: After training our model with over 175 billion parameters, we can finally offer the public a free, slow database that can only be accessed manually from a webpage and is likely to yield completely unreliable information!
Me: [uses it every day]
but can it run crysis?
You fool, you should use THE ONLY database that will ever be needed ever : SQLite !
@@monad_tcp Yes! I hate databases that can handle more than one transaction at a time. What are they trying to prove??
@@JesseHughson that even with one transaction per second that's a better database than bitcoin ?
GPT please hallucinate the next hash
Even more amazing if you think that while he’s doing this things for you, he has 100 million chats aside going on, giving each one of them amazing answers like he does with you. Imagine the power.
Good point, very impressive to consider the scale it operates on
"she' ;)
it's not like one computer running that services all the chats. It's pretty much the same like everything else on the web - there's many different machines running the same program to service many different users at the same time
this is..amazing
Wow amazing !
Thanks - glad you enjoyed watching!
absurdly cool
This is wild!
I asked it: If Earth was sentient, would it consider humans to be parasites? ... The answer was excellent and actually surprised me.
Incredible! Mind explaining how is it practically possible simlating sql server db? you think, it's really using/accessing sql server db at the back end?
Good question! It’s part of how machine learning and training an ML works. ChatGPT must have been trained on SQL literature, code and samples. It’s not really accessing a database in the backend. but during its training it must have seen many SQL statements and outputs, so that it can simulate it. Hope that make a sense.
@@embracethered It looks like its providing you with MS documentation but with some "human-like" or 'teacher-like" additional explanation. Also it adds real output from SSMS results field, which should be part of SQL server tech spec. If I make for example a word doc for my front developer, I write very same statements on how procedure works and how to execute it, with examples. Probably the chat is automating the tasks that it has been trained of. Just getting the answers much much faster.
That is unbelievable!!
Thanks, I think ChatGPT is really cool!
If you haven't prompted it to list any databases, you can make it make up stuff, and its kind of cool
Good point! It’s really good at making things up. Thanks for pointing that out.
Mega. I have asked it to knock up some boilerplate python to confirm I'm on the right path and / or introduce the odd library and function to me. I specifically went in search of ChatGPT SQL because I didn't think it'd be any good at it. Crazy. I suppose it'd be difficult to ask for help with real life problems without first having to explain all of your existing tables to it though.
Definitely worth exploring more. I'd imagine you could tell it the DB schema, and it will be able to create a proper DAL for it in C# or Golang for instance. It's actually something I want to explore more - e.g. have it create a three tiered app basically.
I’m curious to know whether ChatGPT would also respond correctly to syntax errors and object reference errors, provided the context you built for it. Moreover, could it to some extent also simulate SQL server’s execution plan generation, given a query?
For syntax errors it basically ignored them initially and it tries to figure out the “intent”. Like for simple errors, such as mistyping a keyword it will just auto-correct and keep going. But, if you prompt it to flag syntax errors it didn’t run them, but gave an error and explanation what the issue is. For totally invalid queries it replied with “please provide a valid sql query”. Learning and experimenting prompting is really the key to get the desired results with ChatGPT.
One of my first experiments was to use it to code review existing stored procedures for issues - that also worked well. At least for spotting SQL injections.
Hope this helps! I haven’t looked at a query plan prompt at.
@@embracethered thanks for the reply, and by the way, also thanks for showing such a truly interesting use case.
@@embracethered are you saying we finally have better error messages than "There's a syntax error on line 42"?
@@rioghander2te Try it out for yourself - next time you get a syntax error paste the line it into GPT and ask what's wrong.
This chat bot has a great imagination
- ChatGPT open the door!!
- I'm sorry, Dave. I cannot do that.
👀
Imagine you are my personal door operator...
With this video I've now gone and asked it if it could be a Commodore 64. It misunderstood and gave me an example, but that's led me to providing it corrections to the oversights it had.
First, it seemed to think that lines beginning with a number would execute instantly. That's wrong.
Second, it didn't know that user input was requested with a single "?" on a new line.
Third, it (either didn't know or just forgot) that numerical variables have spaces padded around them in output. So the output of what it gave me would've had some double spaces.
It's ability to learn from corrections is absolutely insane. I never got into the fad until now but now I think I'm a little addicted.
Very cool! Like the Commodore 64 exploration!! 👍🔥
I created a very very simple BASIC prompt, it’s here: m.ua-cam.com/video/zNlmbnbiJTk/v-deo.html
I guess I was most excited that it was able to print the text of the C64 Home Screen!! 😀
It can simulate the output of code too, quite interestingly.
Does it save the session data? Maybe save the table to some individual data store somewhere? Can you go back after logging out and run a query against the data you created?
I've been holding a "conversation" with chatGPT that started with a stupid inocent question and devolved into a battle to try and outsmart it or see if it fails. I posed a riddle to it and it gave me a wrong answer, I simply told it the answer was wrong and chatGPT said "Ok let me do this a step at a time" and it did and it figured out its own mistake and gave me the correct answer and even said it was sorry.... I was like what in the actual f*ck?!?
I'm still extremely amazed by it...
Chat GPT , the soon to be greatest programmer on earth.
If you're a Plus user, don't forget to set the Legacy mode unless it won't work. Great usecase btw
The power of programming isn't the computer, but rather that it's a formal language. This really shows the power of the lack of ambiguity.
this is AMAZING
Thanks for watching! Appreciate the feedback.
lmao...'a hacker has to be #42 of course!!'
Great stuff. subbed
Thanks, appreciate the comment!🙂
@@embracethered thank you for the content. Haven't had a chance yet to go back thru some of your other videos but I like your delivery and content. Looking forward to future stuff too and totally preoccupied with trying to catch up a bit on this whole ChatGPT thing!
i used all database from my uni and put it in chat gpt so i can get a good idea of whats right when reading for my exams worked good
😂 this is awesome !
Thanks for the comment - glad you like it!
I'm a SQL Server DBA. I tried using ChatGPT to diagnose an in-memory table problem were are having in Azure the other day. It couldn't help. In the end we just Googled and found an MS KB article with the fix.
Curious, was the KB article dated after 2021?
@@embracethered - No. It's complicated by the fact that the issue applied to version 2019 but the KB article related to 2017 (KB4052338). As it was the only lead we had, we tried applying the advised Trace Flag in a non-prod environment and it worked (although we then had the database go into suspect mode after service restarts so we can't use the fix). We're opening a support call with MS. I suspect MS will give us an undocumented Trace Flag. With regard to chatGPT, it may be that it discounted the KB article because it didn't officially apply to 2019.
Watching roleplay with a computer gives an interesting feeling.
Thanks for the comment. Yeah, I like how you call it role play - it’s basically what it is! Mind elaborating?… scary, exciting or just weird?
Like ChatGTP is a grown man at a tea party.
I've been stumped on how to use the software but this is it: the pinnacle of interactive entertainment.
You bring up a really good point with interactive entertainment. I haven’t thought of it that way, but must be part of why it’s so popular.
this is mind blowing.
Thanks! Glad it was interesting!
December version was the smartest one
Wow, I may need to start looking for a new job. LOL Thanks for posting.
now imagine this but with gpt 4, more languages, more knowledge and (if we get too far) quantum computation!
Exciting times ahead!
I hope chatgpt will be able in future to simulate photoshop, autocad, illustrator etc. This will give us text-to-graphic/cad/svg
Now i need a standalone version of ChatGPT 😀
We all do, but even without knowing what your PC is like, I know with a high degree of certainty that you can't run it. Not in real time, anyway.
According to chatGPT you'd need at least half a terabyte of RAM and at least 64 cores at 2 threads per core.
Estimated cost would be anywhere from $10,000 to $50,000 but I think you'd have to be pretty savvy to do it near $10k. And then, of course, the code, which isn't for sale at any price.
You could just buy the company if they were open to selling, but that'd be at least thirty billion and probably more like forty. And I they'd be foolish as hell to even entertain an offer at this stage.
Haha, that's brilliant! :)
Glad you enjoyed it!
@@embracethered Thanks for tip! :)
You could effectively design a database then ask for a transaction-wrapped output of the entire database (structure and data) to then import it into a real database lol. I don't know why that would be much better, but there could be a use? I certainly wonder how much memory it starts to use for each conversation, especially one where it's taking on a database lol.
Wonder if it could create a database diagram in some form of svg or xml that can be shown as image if you provide it an entire database schema.
even more fascinating: it uses the function without proper storing it ;) i mean it was just giving an example
This is crazy :))) !
Would be interesting to see if you make an syntax error, will it see as error or will it query as it would be correct.
I have some info on that, so if it gets the intent of what you want to do , it will provide a result. To trigger syntax errors, I had to explicitly tell it to be picky about typos and so on!
absolutely insane
at 3:30 would it have showed two separate commands for your inserts if the first command didn't have a semi-colon after it? Playing with alter database and something like set compatibility_level would have been fun too.
Good question, I don’t think the semicolon made a difference. It technically are two statements in a single batch job - so the output messages are actually accurate I believe. Definitely worth to explore some of the possibilities in more depth I’d say. Fun times.
Awesome.
Thanks for the note. Appreciate it!
I think you can get it to write the output in columns, might have. to describe it in the initial prompt.
Yes, by refining the prompt you can get nicer output. I remember using the words “code style block” in the prompt helped, and as someone else suggested “markdown” can give better output. Prompting AI seems like the new skill to have for the future.
@@embracethered I asked it to revise a blog post, and it was loosing all formatting. Then I was asking it to output it with markdown formatting, and, to my surprise, it outputted it nicely formatted. That's when I realized that the web frontend is automagically converting markdown to HTML, and ChatGPT is probably not aware of it so we have to tell it to use markdown formatting in the response.
In the future:
"Yo, android B-#345, can you compile this code for me?"
"No problem homie"
It gets even crazier. You can tell ChatGPT (using a simple sentence) that you want it to insert a ton of fake data into your imaginary users table. It will then do so, and you can continue "querying" as if all that work had been completed.
I've heard ChatGPT described as "just" a text prediction tool, but this is legitimately amazing. In order to hold this conversation with you, ChatGPT must be maintaining some sort of context that describes the state of the hypothetical database. I'd be fascinated to learn how that's done. Is there some sort of SQL plug-in that understands how to mimic a relational database ? Or is there something more general going on ? Could you actually teach GPT to write and execute code in a language it had never seen before ? I almost don't want to know the answers, because it would spoil the magic. Incredible stuff !
Yes, you can teach it stuff it doesn't know. There was a blog post where someone together with chat-GPT made up an alien language, set the grammar, made it construct a dictionary and then construct sentences. It's wild!
The wild thing is, it's doing all of this basically by prediction, it can't "execute" anything, so it only works with the text you send it and the results from previous prompts
@@geli95us The previous prompts are the DB though. All the information required is in the previous prompts.
jesus christ, that's impressive
amazing.
Thanks for the comment. Glad you like it!
Interesting :)
I wonder if you try to input a large data to it, what would happen ?
"Chatgpt can't replace content creators"
Chatgpt: "and if we run the select statement, we can see that it did infact update as opposed to inserting a new row"
It’s pretty impressive indeed! Thanks for watching and leaving a comment.🙂
I told ChatGPT: Imagine you are a mouse.
To cut a long story short, a Jedi appeared and saved the mouse from a blind cat then adopted it.
Fun story!
lol
could this emulate an asterisk server? maybe simulate calls or something like that on VoIP?
that's really interesting. A big part of my job is reverse engineering databases. I'm gonna have to give this a try to see if it can help there.
It seems as if they did have a lot of SQL training data for GPT-3, so it might indeed be helpful.
I can’t really follow all the negative voices which just think it’s text pasting, this is so mind blowing deep „understanding“ of the concepts it learned.
Right? Also, I drives me crazy the fact that these haters are completely missing the point, that he simply asked chatgpt to create the SP for him... How people are not noticing that? Like... lol?
It has no understanding. If you know how self-attention-based models work you will realize that it is something totally different. In the end, it only estimates the next character based on the Input. Because These systems can have very Long Context, you believe it's intelligence. It's not, Like far away. Semantic reasoning with Common knowledge is the holy gral and No one discovered it yet. Deep learning is fundamentally Limited.
Here is another cool prompt to play around with: "imagine you are a bash shell. respond only to my commands as a bash shell would without any additional output"
Oh, I've seen that. Even went down the rabbit hole of simulating the running of wget and installing and running chatGPT in the simulated server.