Mind-Blowing Automation with ChatGPT and Open Interpreter - This Changes EVERYTHING!
Вставка
- Опубліковано 9 вер 2024
- Using the Open Interpreter, it is possible to give ChatGPT access to your local files and data. Once it has access, automation becomes a breeze. Reading, writing, summarizing, modifying, renaming, all just by asking nicely! Open Interpreter uses ChatGPT to write code to fulfil the requested task and then runs that code to do it. Brilliant!
---
Let Me Explain T-shirt: teespring.com/...
Twitter: / garyexplains
Instagram: / garyexplains
#garyexplains
Doesn't seem like a good idea 😅
Yeah right, I would def consider air gapping that computer
@@donkeypoopdragoncanu help me understand ‘air gapping’? Is this security like internet block. Seems almost impossible to truly isolate for privacy, but I am open to new concepts.
@stevegyro1 so my Le Potato computer is air-gapped because it doesn't have wifi and there for cannot access the internet without a ethernet. It is completely separated from the internet unless someone were to plug in an ethernet cable
@@stevegyro1ofc it can be done Gary you just have to build the pc from scratch with no Wi-Fi card
Just load it up in a sandboxed VM.
As long as everything runs completeley locally, this is really cool!
Having this containerized without network access would be better
Interesting. Why?
@@GaryExplains privacy being the main issue. I wouldn't want my files or even a summary of my files going across to some server.
Well it depends on what you ask ChatGPT to do. Yes for things like summaries, then the text needs to be sent to ChatGPT, just as if you were using the web version. For other tasks then the data remains on your machine. As I said, you can run Code Llama instead of ChatGPT then everything stays local. Of course getting it to access the internet, like I did for the weather demo, is an essential productivity tool, IMHO.
@@GaryExplains fair enough for the weather usecase, yeah using llama would be a good tradeoff I could find reasonable here. (but I'll still avoid the usecases which require internet access)
The most productivity I was able to think of while hearing about this tool was that, theoretically if it had the option to go through my local notes and allow me for a knowledge lookup. Basically like we question chatgpt but for looking up on my notes. That would be an insanely good workflow!
Tried this. Man... We are so close to get something like "Jarvis" and "Friday" in Ironman.
Couldn't pay me enough money to give that thing access to my data. I love the local options and have been working on designing a dedicated box for it.
Nothing is so "cool" that it's worth that level of privacy invasion.
What privacy invasion are you seeing here?
@@GaryExplains You're giving this thing apparently unfettered access to run code (generated and downloaded) locally and access your data. Is that summarizing the pdf locally or is it sending it over the wire? Is it logging those exchanges for its own enrichment?
It isn't unfettered access, you need to confirm all the code that runs (unless you explicitly set it otherwise). The summarizing service is the same in terms as privacy as using the web UI to do the same thing. Don't use the summarizing service if you don't want the data sent. But of course you can also run a local only version with Code Llama.
How about no
Why? Did you actually watch the video?
I would perpetually live in fear that that it would 'hallucinate' I wanted all of my files securely erased!
So insightful. Thank you Gary!
I think this is the one thing that I will do myself. If I stop taking care over my files and data on that very basic level, I fear, I might completely lose touch with what these files even are.
Then you will have to ask chargpt what those files are 😂
Maybe that's the next level, not directly engaging with files, I mean they had the same anxiety going from assembler to compiled languages lol
This is a game changer...Thanks Gary.
Awesome! Thank you for showing the possibilities! Looking forward to a gui version of this 🤤
Hi Gary, I'd say very impressive but that's an understatement. Time for a practical arduino video example? ChatGPT writing code for an arduino demo, something along the lines of Star Trek, just a thought. Thank you for your time, videos AND enthusiasm.
All the other UA-cam videos about open interpreter jump head first into the program with out explaining what exactly open interpreter is. You did a much better job than the other UA-cam videos on this subject 👍.
Thanks, I'll have to look at that Liam's chat program video you mention.
This is amazing & mind blowing!!!
Jarvis is here!!!!!
I thought about mentioning Jarvis but I thought that maybe it was a bit of overkill! But basically I agree with you!
Crazy ! I have a question : when you use chatGPT, your data are transfered to openai, so there is a risk, no ? As I work on companies documents, I should instal llama localy ?
FACT: It is extremely foolish to let an online AI language model openai or otherwise have direct access to your files and data. Too early to know the risks? but you probably already have a idea of what is possible. Instead be safe use at a local large language model on your own PC that is not leaking your data to a corporation that you have absolutely no idea what it is doing with it.
FACT: Extremely foolish is an extreme exaggeration. You decide what you ask open interpreter to do. You decide what files you ask it to open and read. You even get a prompt asking if you want to run the code. You are in control. Yes you need to be aware, but the term extremely foolish is extremely naive.
I can't imagine having this without a version control system in place (possibly automated). Imagine prompts that are wrongly interpreted as "delete..." or "encrypt..." or caused by hallucination.
I would hope you have proper backups with or without this.
You have a point
Funny you mention talking to it with speech-to-text, I did just that in a experiment the other day.
Given the current title a lot won‘t watch the video and just start a controversy about sharing your personal data with Open AI.
Having this sort of phyton script based personal automator is a nice idea. The use cases will be bound by the bandwidth and the context window size for this all text based approach.
Gary, this is Amazing, revolutionary, the potential is staggering. We are on the brink of Star Trek like technology, you were 100% right to be extremely excited. Thank you again.
At this level, and with the kind of tasks you just demonstrated, it is probably faster to just use your command line manually. but, I believe, not for much longer - this is a glimpse of the future.
No not really, I doubt I could make a graph of those foods in less time it took me to type it.
@@GaryExplains Not in Python, but directly in the same Excel, it would probably be a tie. Rename and counting tasks would be faster by hand. But i *do* get your point, of course - this _is_ exciting.
I think you also have to consider that you can do this while being completely rusty or perhaps not even knowing how to do these things yourself. Maybe if you're in practice you can do it faster, but there's no way you're gonna research it, learn it, and actually do it yourself faster than this thing can interpret natural language.
@@xxportalxx. Yes, you have apoint here.
If it is soley contained on your computer/server, maybe. If it has access to the internet, absolutely not. Giving ChatGPT access to anything on your side of your firewall is asking for trouble.
Why. Has ChatGPT been known to start hacking in response to a prompt about renaming files or similar?
@@GaryExplains Given enough time, Not that I know. However, given enough time, ChatGPT and similar programs will be used ,by unscrupulous people/governments, to do things that ought not be done. It's only a matter of time.
I think it might be the next new feature in upcoming window through bing...
Gary, one thing I'm not sure about is best practices for multi-step "Command" do you need to do it step by step or can it handle a single complex prompt. Example,Rename those JPG. Resize them to "X", Upload the files to this website, Create a text file of each JPG describing the main subject. Where does this tech break as you give it a more real-world task. Thanks.
I wonder if this is the type of excitement the people of Star Fleet had when they "computer."
The last time i saw ur video was probably on Android 8 lol it's been years
Very cool to see u making ai content and very very surprised to see u in a search for open interpreter i used to remember seeing u only in the context of Android
This is nice to see
The only danger AI presents to any human being, is the danger of being left behind if you don't get on the bandwagon.
entire video is just so fun to watch, many small software on our system and those companies wiil be out of business soon :D 😄
Mind blowing stuff
hey #Bing why do you need my PIN & Mother's maiden name?
well that scared the hell out of me. moar plz, AI definitely needs more discussion in the general public
edit: good video
I am waiting for a fully vocal interaction where I can make chatgpt use the voice of Majel Barrett. It is only then will we have arrived
I want K.I.T.T. from Knight Rider 👍
I want K9.
It said the column calories is measured in grams. Not very smart, is it?
Excellent. I'm going to give it a whirl in my learning of doom emacs (coming from neovim), using kubernetes, docker, cargo/rust, and ansible. Ambitious. As my mother would say looking at my dinner plate, "Your eyes are bigger than your stomach." ;~)
What will be cool if the open interpreter application had vision capabilities, where it could look at the image and rename it accordingly.
NEXT: random, UA-cam Commenter gives next idea for video
I believe that you can voice activate it already using standard dictation in Mac OS...
Good point! I just tried it and it works great!!! 😁
I wonder if you could use voice to text software or the native stuff on windows and speak to it that way
Oh wow a new way to completely bugger the metadata on mp3 files
MP3? Really?
@@GaryExplains could be interesting. Tag all my music with the correct discogs release.
@@GaryExplains zero chance I would take a risk like that
You heard of backups? Not hard to copy all the files to backup, take the risk, and then if it doesn't work, restore the original files.
@@GaryExplains we spend the rest of our time on security and then hey install chat gpt on your own network!
Please post the link to the tool you're describing in the video description
I up graded my Mac OS and my driver to my Light Scribe DVD burner on longer worked. Could this ChatGPT fix my driver? So I could again burn pictures on my DVDs. Fixing software would truly be useful.
can you still use this tool with chatgpt 3.5? the -f option doesn't appear to work and it seems I'm only able to use chatgpt 4
Giving ChapGPT access to your files and data … hmmmm
Nice
very interesting!
Truly amazing. Mind blowing. I even never ever dared to dream about it. WOOOOHOOOOOO. I need to get that running on my PC.
That's next level AI usage... I would be a little worried giving GPT full access to my local data. In deed I would not be allowed to provide it access to my customer's data.
But in deed I did almost exactly your mass-renaming example yesterday with GPT-4 (in Bing) - just the semi-automated version, by letting it generate a small python-script to remove the first few digits of every file in a given folder... not complicated, but GPT-4 wrote the code in 1min, where me - writing scripts only every other year - would have taken several try & errors & google-help to get it running.
Oh wow. This is really a game changer. Can it sort out files so that all images are put into an mage folder?
For sure
But doesn't that already happens in Windows 11 and with GPT-4?
Was a blind, or did you forget to attempt these somewhat "vague" messages? like "make a file with some info about a famous actress" or something
And also refer to a previous command you told it to do
i will try makefiles
The real question is, who else have you now given access to your files and data? As we move into the future, the more we rely on there things and less on ourselves, the more likely there will be individuals willing to take advantage of that fact.
Did you watch the video, because it seems quite clear that access isn't granted to anyone else.
@@GaryExplains Access can be written into the code, similar to the way it has happened on some apps today. Is that not possible?
So you are suggesting that if you use ChatGPT to write some python code to perform a task like I demonstrate in the video, that it will deliberately include code to upload your data to some secret place? Why would it do that?
@@GaryExplains ChatGPT doesn't exist in the ether and needs code for it to write the code you to write with let say Python. That original code can be manipulated. But look at the data we already give Google, Apple, the Chinese and others, so why not?
Sorry, I am still confused. Any online service including Gmail or Facebook could be manipulated to do nefarious things. Even Microsoft Word, or the new Starfield game. I am still unclear on how you think this will be any more likely than any other app or service serving malicious code.
Trying to set this up, got credit and all on my account but it says gpt-4 model not available.. :/
I assume you cut and paste your key when it asked?
@@GaryExplains Yup turns out i just needed to wait for it to notice i put credit in. 👍
Crap Im studying industrial automation
feels like a long-winded way of executing simple commands. granted that it knows to install the right utility, but its not really a new way of doing anything. it may be beneficial to a non-geek, but the tasks you showed required a thinking that comes with knowing what you can do with a command prompt. Once this thing can move into the realm of automating GUI stuff between a set of general apps ... by all means do wake me up.
I'd rather ask bard to summarise an RSS feed.
Yeah, I don’t think this is a great idea.
Why?
What are you smoking? 😅
Eh?
Cool. But I will wait for MS Windows with integration of GPT. Those AIs are so good.
how much ram do you have on your pc? i am thinking of upgrading RAM on my laptop from 6gb to 64gb. i am planning to upgrade because i want to use ai tools like stable diffusion, offline llms and other ai powered apps. do you think 64gb is a good choice or 32gb ram will be sufficient? Share your valuable insight.
Laptop that starts with 6 GB of RAM doesnt sound like a powerful system. Not that more RAM won’t be a good thing but possible it won’t really even be able to fully utilize 32 GB of RAM thereby I seriously doubt 64 GB will even be useful.
GPU is also what is needed to best use these kinds of tools locally, where it will be limited what you can do with a CPU.
Yeah you need some pretty good ram minimum 64 ideally 128gb ram for local llm
@@AndersHassyou must not be in tech because anybody that has used local llms know that lots of ram is needed to run these tools
@@perc-ai I didn't say you don't need a lot of RAM.
Do you really think a CPU that came with a laptop with 6 GB of RAM can utilize so much RAM?
Usually you would use something like a threatripper, Xeon or Epyc to utilize 100+ GB of RAM.
@@perc-ai I already installed 64gb ddr 4 3200mhz with with cl 22 timings
Gimme lots of trojans plz 🤭
Is ChatGPT known for writing trojans?
@@GaryExplains I'm referring to prompts injection. Also privacy, I know nobody care nowaday.
@@Eratzer Prompt injection from where?
this'll become a necessity to stay competitive in the workforce within the next years.
if that is not the base of windows 12 microsoft is a lost cause at this point
The cost for millions of users is probably the reason you won't see anything that impressive, I'm using the beta of windows co-pilot and it can change your desktop background or turn on dark mode and that's about it.
You: wow! Me: does it have garbage collection? In the AI case: does it verifiably have a theory of the world? As for the renaming stuff: I can do it myself with a bash for loop, why should I do it with the same number of letters in plain English not being secure on what I get? You certainly got a bash command, but you had to control it yourself and accept it. Why then didn't you just write it yourself? Sorry: your use cases demonstrate that you need your own theory of the world to verify the proposals, and still then you are not quite in full control, which is the exact opposite of what a responsible system administrator or programmer would wish.
Noice
The paranoia on display here is slightly bizarre, Windows already has access to all your files, so do all of the programs you have installed to manipulate them. Why trust them and not OpenAi?
Exactly
Arguments i can think of:
Newer Tech, more things we are not aware of.
More security holes maybe , also through "wrong assistant software" that someone uses who knows less about security.
being too easy on new things and too quick is often instable (meaning uncontrollable things could happen easier).
But idk much about all of this ^^
I think its because Open Ai is newer mayb@@GaryExplains