How to let GPT control anything & 10x powerful | 8 mins tutorial about GPT funtion calling

Learn how to use GPT function calling to turn GPT into an agent that can do wide range of tasks;
- Follow me on twitter: / jasonzhou1993
- Join my AI email list: www.ai-jason.com/
🔗 Links
- Github link: github.com/JayZeeDesign/gpt-f...
- RapidAPI hub: rapidapi.com/hub
⏱️ Timestamps
0:00 What is function calling?
0:53 Tutorial overview
2:18 How does function calling works?
4:31 #1 GPT access live stock data
9:09 #2 GPT access Airtable
10:56 #3 GPT controls smart home
👋🏻 About Me
My name is Jason Zhou, a product designer who share interesting AI experiments & products. Email me if you need help building AI apps! [e-mail]
#gpt #langchain #autogpt #ai #nocode #tutorial #stepbystep #langflow #flowise #gpt #lifx #openai #smarthouse #smartphone

[@gustavobenites5438]

Best YT Video on the new Functions Calling from OpenAI. Congrats!

[@AIJasonZ]

Thanks man!!

[@JoaquinTorroba]

Could you create a video explaining how to use "function calling" with Flowise or Langflow? Does it make sense to use "function calling" with langchain or they have a similar role? Thanks Jason as always 👏🏼

[@chiragthesia2065]

Dude your videos are amazing! Keep at it!

[@AIJasonZ]

Glad you like them!

[@TeamUpWithAI]

You're such a good teacher! I can never do and talk at the same time about what I'm doing, so making tutorials is a struggle. I admire your skills :D

[@DRINOMAN]

best video i’ve seen so far on the practicality of these function calls nice addition with the lights.

[@celsospencer416]

I extend the highest vibes to you Jason! Excellent train of thought, you are an AI whisperer. Email coming your way right after I consume the rest of your awesome content. Greetings from HM.

[@clairewhite3461]

OMG! It's so powerful. Great video, thank you!

[@adamrodriguez7598]

Thank you!!! Love these videos. Super informative

[@losokos5558]

This is AMAZING so much potential!

[@jaysonp9426]

Dude...this was the best function call video I've found. Thank you so much! Subscribed

[@chaseclingman]

Very helpful, thank you Jason

[@mauritsotrolado7857]

Wow, this is super cool! Nice work bro.

[@rezomomo]

love your videos dude!

[@shivamroy1775]

Great content, please continue making such informative videos

[@LeighBriody]

Finally a good video that explains this well , thank you

[@derekcheung2598]

Outstanding work Jason

[@kevon217]

Great video!

[@hchentw9154]

cool thing, thank you Jason. Great videos.

[@ColinTimmins]

Very good content. Instant sub from me, I’ll have to look at your other videos for sure. I have made some amazing stuff but I’m still learning how to piece it all together. 😊

[@sayantandas7544]

damn man!! really awesome video!

[@tugrulpinar16]

awesome content, keep it up!

[@rory1653]

that is pretty friggin cool

[@test12382]

Dude Ive been on this channel for about an hour now

[@colmxbyrne]

How do embeddings and function calling work together? Can you create a tutorial on that?

[@kamalkamals]

this morning i create a small app with new features functions + langchain, it s very good idea because that s help for many cases like websearch ...etc

[@user-oq4ok5cq4v]

wow what a cool feature, thanks a lot for the tutorial i'm going to test this with my new RGB lights and a few other stuff. Also, can you create a video using the Ai gf concept but more in an assistive way integrating with a telegram for example?? it will be very useful to have that lovely assistant on the phone if I need something like searching for something or playing music etc.

[@jdoejdoe6161]

can we also leave all the irrelevant outputs and show only the desired output in well formatted state?

[@jdoejdoe6161]

Can we make this prompt work? "What are the five stocks with the biggest price movement today and what are the latest news about these stocks that might have caused the price movement?. can we write the output in a normal table format instead of airtable? Thanks.

[@kevon217]

Wondering what errors you’ve run into and if you need to consider any particular try/except blocks.

[@mbrihoum]

Hi Jason, I'm interested in learning more about the possibility of creating an app that links directly from Excel to Thinkorswim (TOS). As you know, Ameritrade offers real-time quotes and news, and I think it would be very valuable to be able to access this data directly into Excel. I'm not familiar with both VBA and Python, did you explore the option of linking excel directly instead of using airtable?

[@AIJasonZ]

Great suggestion! I’m making a video now in terms of integration with google sheets, will let you know once it’s done!

[@pleabargain]

The top moving stocks changed from COHR at 8:13 to NEXT at 10:36. Can you please provide an explanation? Is there a way to force a link to the news source for verification?

[@OryginTech]

Price movements change every second.

[@holistic_hass]

Great video as always Jason! PS: Check your emails ;)

[@jadson.empreendedor]

Am I mistaken or does function calling have the same effect as me making a query in the database via Python, for example (SELECT * FROM table), returning this query as JSON to GPT-3.5 and asking for it via role SYSTEM , USER etc to know who my top customers are? Do you know if it's the same as creating a call function for this, Jason? Thanks.

[@bobuputheeckal2693]

Hello Jason. Nice presentation and use case. I have two questions. 1) do you know how to get todays date? I have a case where user want to check availability of a service for tomorrow or in future date. 2) in function calling, while describing function, can we default an argument parameter if it couldn’t be determined by chat gpt

[@reinerheiner1148]

Just use the same principle to give gpt4 access to a function that returns the system time.

[@Meza201]

ChatGPT already knows the date as it’s provided in the hidden prompt/ context/system message. Ask it the date, it will tell you. And if you ask it how it knows it will say it’s from the context you gave it previously.

[@Pork-Chop-Express]

Can you show this using the S&P 500? And then provide the code?

[@edwardyb8970]

Where to get types, parameters and objects from? Are they in any api documentation?

[@AIJasonZ]

Oh yep! They normally lives in the documentation of the api endpoints you are using :)

[@nat.serrano]

great content. how does GPT knows what function to call? that's what's not clear to me. how does it decide to return text or call a function bsed on the auto parameter? just based on the DESCRIPTION of the function? what if the prompt by the user is bad (or doens't fully match the description) and doesn't pick the function based on the description?

[@AIJasonZ]

I think they are using a prompt prob something like “you will decide if any tool in the array […] will help move towards the goal user intended, if so, return xxx” You can try similar prompt with GPT and I think it should work

[@IceMetalPunk]

Yep, it's deciding on its own based on the function name and description you provide it. If you set it to "auto", that is; in which case, if the user's prompt is exceptionally bad and it can't figure out what it needs to do, it'll just return text and not ask to call a function. You can also specify a particular function name instead of "auto" and it will *always* try to call the function you told it to (which is good to let GPT figure out what parameters to pass).

[@nat.serrano]

@@IceMetalPunk thanks, make sense

[@ko-Daegu]

where do we get/control the functions names I guess prompt engineering right ? If I want i can tell gpt model to use specific tool name when needing to do certain thing or do something using tool called cuz I'm grapping the name right ?

[@AIJasonZ]

Not sure if Im getting your question correctly - but you will define an array of function descriptions, where you will define the name of function and pass it to GPT

[@Mzulfreaky]

More pls

[@maxnakama21901]

I am still in the middle of watching the video, but I was wondering if the "functions" that gpt calls have to be real code functions or they can be "functions" that aren't actually code, but rather a text based version of a function? Like get_stock_review; does it need to be real code or can it be just text? I am explaining poorly.

[@AIJasonZ]

You are spot on, it doesn’t need to be real function! I’m actually making a new video to use function calling as a way to retrieve structured data, will share once it is done :)

[@chivesltd]

can you do a video that uses the function calling with AutoGPT?

[@AIJasonZ]

This is a cool idea, let me think about it’!

[@pleabargain]

Will this code work with gpt3? And you misspelled function.

[@tomich20]

Can we use this by creating a chatGPT plus plugin or its only available through openAI api?

[@AIJasonZ]

I think function calling is actually the backbone of how plug-in ecosystem works! (So chatgpt using the same function calling to decide which plug-in to use)

[@royalauclair]

It's cool. And nice showcase, but at same time you could have just used the functions you made to do that exact thing without needing chatgpt at all. The LLM usefulness here seems pretty limited and overkill?

[@sl554]

You're missing the point of what the AI is actually doing here - the function and arguments give the model instructions of what it should do with the provided input. It's a better / more structured way of providing a list of things you want back. For example, you could create a prompt that said something like: "Give me a list of: 1. priority 2. Order amount 3.... Etc..." And get mixed results. Or use the "function calling" fine tuned model to give you a response in a predictable format that can easily be parsed!

[@ihateorangecat]

Hi! Am New to Open ai I want to know what's different with Auto Gpt and this "function calling" thing? Please someone help me understand.

[@IceMetalPunk]

AutoGPT is just looping GPT's input back on itself to create an infinite loop. The ability of it to do things like browse the web or spawn new agents is just provided by the framework, and is parsed out of the AI's English responses, which means it's not super reliable. (I.E. if the AI responds in a way that it can't parse, it just errors and tries over and over until it happens to work.) In this case, the AI itself is fine-tuned to understand how to call functions, *and* it allows you to specify your own functions to the AI, making it much more robust and reliable for general use, without needing a loop-and-parse framework.

[@jdoejdoe6161]

can we add sentiment analysis to this?

[@AIJasonZ]

Yes! I’m actually making a new video about it :)

[@yogiwp_]

Does this work on 3.5?

[@danielsevilla5216]

Yes. You have to specific the model as shown in the video. Turbo-0613

[@blazi_0]

You have typo in the title

[@zac1427]

wooow，真牛逼啊

[@AylaCroft]

Subscribed and followed you on Git & Twitter

[@kelcitangri4429]

*promo sm*

[@Ray-mf6vx]

We don't need Langchain anymore

[@atomicHabbit]

elaborate please.