BabyAGI: A Real First Test
Вставка
- Опубліковано 25 кві 2023
- A first walkthrough of BabyAGI - starting at zero, speedrunning through install (pretty easy) to first test of an actual use case (comparison shopping) to modifying the script to unlock the hidden memory of the Autonomous Agent (not that hidden - it’s ChromaDB). All wrapped up with final impressions and API usage (not bad).
As always, please comment with any idea you have to run BabyAGI through the wringer - or any other Autonomous Agent.
Repo for modified code: GitHub / ContextFound / babyagi - Наука та технологія
Hi, I recognized myself at 6:35 when you said "you made me do it, I had to open up the code"... I'm a Master in arts and French letters; when someone ask me to read a text of his own, as long as I can close my eyes on the mistakes I see, I disallow myself to start rewriting the text the right way. So when you say: "You're done, you made me do it, I had to open up the code", I understand exactly all what it mean at the same time.
Despite our opposite backgrounds, we both run in the same rabbit hole (i.e.: I tested AutoGpt yesterday and thank you for sharing the test you did with it in another video, my conclusions about it were very similar to yours, ). We should find a way to get in touch and have a talk. Do not hesitate. Have a good day.
And I was being reminded how good we have it today even with narrow AI as I watched the video on your channel with auto-translated captioning :)
There is a channel that has "ai agents". Guy with a beard.
@@raresouls9897 I guess I'm the the guy with a beard? What channel?
Great videos. I have been doing similar experiments. Amazing how quickly these implementations are being developed. What a time to be alive! 😂🤖
"Squeeze those papers, fellow scholars" :) It only seems to be accelerating!
I'm looking at your timelog. No doubt you probably don't have to show up at work. For your health and sleep I recommend following the protocol of lighting, get to bed by 10:00PM and up at dawn, get that morning IR and again before dusk. You will feel much better and you won't look like a Japan arcade resident.
Great pile of dog food BTW. I need hand holding - but you came pretty close.
This is so good that you do this. So I just see the results. Thanks
LOVE THE VIDS I hope you make more. Also, please try the newest versions of Auto-GPT and Super AGI. commented on your newest vid to show more opinions.
Thank you for the niche insight! 🙏
Good video.
This is amazing
What's most stiking to me is the contrast between the complexity and capability of the projects with their "simplicity". It is wild how approachable coding these is.
So what could be some of the experiments we could do to help it not get in death loop, would it help it is explained what it was doing and why and have ability to intervene and ask it to do something else like read a paper which gives it new things to try.
Added to my list of tips to try. I think the next generation of more successful bots will be based on a more conversational interaction - I'm trying to bake a "gather more information if you need it, just keep running if you don't" approach into a tool.
I would love to have this on some real-life use cases that we have at one of our productivity software companies.
I'm excited for personal productivity applications of LLMs, and am working on an experimental design myself. Whether it's Microsoft or some solo developer, this will change the game for productivity.
i think it needs a " that will do" function to complete the task
Agreed. I conceptualize it as a "wrap it up" command.
Try agent agi
That sounded familiar, so I went looking for it to put on "the list". Are you talking about reworkd/AgentGPT?
@@ContextFound yea