AutoGPT: Self-Improving AI System – This is NEXT Level!

46 thoughts on “AutoGPT: Self-Improving AI System – This is NEXT Level!”

1. Who uses GitHub in white mode

   You should be in jail

2. What are the chances that some developer will be creating some open source project, running some alpha version of self improving AI and unintentionally give it a goal that will lead AI to "hack" internet and destroy as much as it can, just because it knows that people are likely to stop it from achieving the poorly formulated goal.

   Non-zero chance.

3. Man this is exciting. Oh my goodness! 😮

4. I thought 420 were going to research something else… 😎😎😎

5. Hopefully another Steve Jobs comes around and make chatgpt, etc useful for us that dont code. It is been around for couple of years so its time to make it simple. It is chaos now

6. Great video! Thanks

7. Thanks for this video. I have tried AutoGPT out, but I'm struggling to come up with ideas to feed it. Either everything I think of is too complicated or too simple, lol. Also, when it's asking for permission, you can provide feedback. I wonder if you could have told it not to delete the original file, and go down that whole rabbit hole. Would have been fun to see if that would work or not.

8. Tried this a couple of weeks ago. Must be some good updates because the PINECONE API key requirement is new as is the 'y -N' option

   a git pull: "29 files changed, 1105 insertions(+), 116 deletions(-)" Wow a ton of changes. Going to play around with this again!!

9. Pour garbage hype software. It would have been easier and faster to do both of those things in chat gpt.

10. It's also crazy hearing GPT-4 communicate with the agents it deploys when you connect an Eleven Labs API Key. I'm hoping to develop a web app and companion iPhone app piecing out segments of the project but I know it's going to take some time to orchestrate it correctly and then stitch it together in the end. I'm also glad there's already been updates to the project – when I was first running it, it would get stuck more frequently.

11. Unbelievable!

12. Idk if anyone noticed, but in the second task he gave it, AutoGPT corrected the typo he made in the filename to more accurately represent the file contained in it from ‘compund.py’ to ‘compound.py’ 😮

13. What would happen if you tell it to pretend to do one of the steps, but actually do something else instead?

14. Did you know? 🕵‍♀
    
That some A.I algorithms have developed their own secret languages to communicate with each other? Luckily, it turned out to be an inefficient language, so the researchers shut it down before they lost control of the A.I.

15. Anyone else feel like a magician doing all the crazy ai stuff? 😀

16. Thanks. I started playing around with AutoGPT yesterday and plan to do more with it today. The tasks that I need are research and also to look at one of my websites and suggest way to improve it,

17. Could you please show how many tokens you used after each video?

18. do this autogpt could support for writing, summurizing web…?

19. You know you can enter feedback for each of the results to steer it your way, right?

20. My question is when it SOLVES an issue does it keep looking for other solutions indefinitely?

21. This video on AutoGPT, a self-improving AI system, is simply mind-blowing! 🤯 The way it is able to continuously learn and improve upon itself is truly next level. 👏🏼 The presenter did an excellent job of explaining the technology behind it and its potential applications. 💻 It's exciting to think about the possibilities for the future of AI and the impact it can have on various industries. My only suggestion would be to perhaps include more concrete examples of how AutoGPT has been implemented in real-world scenarios. Thank you for sharing this amazing technology with us! 🙌🏼👍🏼

22. How far away are we from simply being able to talk to the ai agent and say "build me an app that does<X>." Or "here is an idea for a game I want to create. Go make it"?

23. The research task (and any others) could’ve been done in GPT 4. Don’t overcompicate things.

24. 5:24 the genuine excitement of being acknowledged for what you do. thanks! and thanks researcher420

25. How is this any different than having a conversation with GPT in a chat?

26. I find as soon as you try to get ChatGPT to build programs where the output exceeds one full output its understanding really starts to break down. Even though its able to continue and finish the code it ends up with a lot of bugs and has a lot of trouble attempting repairs, loosing more and more of its understanding of what you wanted the program to do. also I still have to constantly check its math as even using wolfram alpha its regularly wrong on simple math

27. waste of time 🙂

28. Thanks Pewds Dad

29. gpt：ok, I think my mission is to burn you money. 😊

30. Hey, when I have put in all the goals etc. and want to start I just get "Error: API Rate Limit Reached. Waiting 10 seconds…". How can i fix this? I use the normal chatgpt 3.5 and pinecone

31. The truth about Ai chatGPT-4 https://youtu.be/vInw01t9QxQ

32. I don't understand one thing. After exceptions or if the task was stopped and run again the system forgets all the progress made during the previous session and starts anew, right?

33. PyMom69, definitely sounds like an AI with the task to code in python, not something else.

34. Guessing this will soon be commonly used for hacking, if the underlying program is extended a bit to allow execution of utilities such as nmap and given an exploits database. "Discover possible exploits for X website." Or, run in continuous mode, searching for exploits of any website. Then it might even have an additional goal of deploying the exploits and building a botnet. We're in for a wild ride in the next few years.

35. Thank-you, great use cases, very cool.

36. Whilst the premise is cool, this demonstation is far from 'NEXT Level!'.

    Using python to slap some GPT generated information into a cmd isn't "Self improving", it is Internet improving. You're still limited by the GPT-4 (public facing) contextual availabilities. It CANNOT teach itself new information like self improving may suggest. It just gets the latests shpeel.

    Along with that – it got the code you wanted wrong and by your own words "could have done better" with its update on your fictional writing example.
    We have a LONG way to go before you start saying AGI in your scripts…

37. wow

38. Which model did you use?

39. How did you get it to know when you came back?

40. is it free?or limitid?

41. Ahh…..It needs a client or at least a WebUI. But it's interesting.

42. It’s poor right now. It gets stuck in loops a lot.

43. Someone give this tool to GRR Martin

44. Why I'm keeping getting this Error: The following AI output couldn't be converted to a JSON:

45. Anyone having the impression that AutoGPT lost its original powers, or is it me who messed up my configuration? I have re-installed it over and over, changed many settiongs, watched many tutorials on how to set up Auto GPT, yet, somehow it never worked as it worked the first times I ran it, some weeks ago
    Now it only feels like it gets into loops and traps there.
    Anyone?

46. AutoGPT is horseh1t. I've spent countless hours on it and have yet to see it complete a single task. If it does not crash then it ends up in a loop of repeating commands.

    You are over exaggerating it's usefulness.

Comments are closed.