Welcome to AutoGPT Discussions!

[Torantulino]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.

To get started, comment below with an introduction of yourself and tell us about what you do with this community.

[maxxflyer]

hi my name is maxx, I discovered AI in 1996 when computing power was lol and data a mirage. I wrote my first neural network as a java applet. I sponsor this project because during the last week I was planning to try build something similar and this popped up.

[barleyj21]

Hi! This project seems just like what I was looking for, but as I don't have access to gpt4 api I wonder if local llms or even cloud based gpt3.5 are able to search internet with your code?

[rvnikita]

It can search internet because it is done by code, not by gpt itself

[Torantulino]

@barleyj21 Auto-GPT now works with gpt3.5 only if you don't have GPT4, simply run with the argument -gpt3only 😊

[mohsen1599]

سلام‌ نام من محسن هست .من در مورد سازه های مدفون شده در زیر خاک دارم مطالعه میکنم.ساز ه های قدیمی که دارای قدمت تاریخی و اشیاء با ارزشی هستند..چگونه میتوان از طریق ماهواره های موجود این سازه ها را کشف کنم

[jknash101]

I think this looks like a fantastic tool. To help structure prompts and create memory, have you looked at pinecone at all? Building dedicated agents like templates could be a really good way to monetize this. I'm an experience product guy if you want to touch base message me on twitter @j_k_nash

[wanzhiyu-admin]

This is really great.

[IBRAHIMtaiyeroheemat]

It'll get more better

[thebestx121]

When would local language models be available for auto-gpt

[TheAIAdopter]

I tried different way to generate revenue. the last one i tried were generating a script to do automated trading. It worked for a while gathering informations, creating trading algorythm and doing backtesting. It seems to worked very well but it always crash before it finish it's job. I feel like it could developped really good algorythm.

But any large project i tried it end up with bug or it is stuck in a loop.
Do any of you been able to make it handle complete project until it finish it's task? I feel like i'm asking too much and it just failed along the way even if it usually seems to be on the right path.

What are your experience with complexe task so far?

[RoyH82]

I'm very curious which prompt and goals you gave auto-gpt to achieve this. And did it use GPT3.5 or GPT4?

[awepthon]

This is so exciting! I am entrepreneur, running two companies at the moment but also have to many projects in pipeline, this could be a chance that I could manage to set them all off!! I am following this with great interest!

[awepthon]

An idea if you want to experiment with a project i have a few that could be used to see if it works :)

[m-check1B]

I am Matej, I also believe that utilizing templates is the way to go for now. I have achieved the best results so far by rewriting all AI settings and initial data according to my goal-plan-projects-tasks-policies-stats-product_definition template. This template is simple and has proven successful for every project in my 30 years of project management experience. The code is currently too immature to function optimally, but it works significantly better when adhering to template rules. The main challenge was getting Auto-GPT to run the template without veering off course.

I think the template step is a crucial element missing from the current model. Once the model has access to foundational files and rules, it can complete highly sophisticated tasks as it gains momentum. I have been using Chat-GPT4 for coding and various other tasks as an assistant, and I am truly impressed with its capabilities. It knows the necessary steps and possesses ample data to function effectively once it has access to Internet search and data storage + shell commands.

Instead of focusing on enhancing the existing AI "brain," more emphasis should be placed on allowing it to utilize available tools, such as search, file reading and writing, and data storage. The core of Auto-GPT should be minimal, avoiding excessive thinking, doubting, and running in circles, as it only serves to duplicate the LLM model, which is already superior in these aspects. The Auto-GPT should simply task the LLM in the simplest way possible, following the steps of the aforementioned template, and only perhaps allowing the LLM to improve the template after gaining sufficient momentum.

These are my observations and suggestions after four days of testing the current Auto-GPT and two months of daily usage of GPT for various aspects of my work. Hello to everyone here, and welcome to these exciting times!

[askivo]

I have been experimenting with asking AutoGPT to complete tasks in as few steps as possible because the overthinking and getting stuck in a thought and then crashing before producing the final deliverables is frustrating.

However, I have have managed to scroll up and retrieve the output AutoGPT gets stuck thinking about while trying to express itself.

It hasn't been the best at presentation as of yet, but there have been times where a task appeared to be incomplete when really AutoGPT was just tripping over itself trying to do a more thorough job than what I actually asked for.

[ghost]

Hello all!

[lookevink]

hey all! great community. I'd just like to point out that to be an AI entrepreneur, you don't necessarily have to be one of these flashy startups with mind-blowing tech. you can commercialize one of the existing solutions and refine them. I'm currently onboarding our first two customers (both pretty big names) for internal/external context aware chatbots. I made the context retrieval variable length, so the intelligence is miles ahead of other ones. look for a need, and create/find the right tech for the solution

[m-check1B]

I made the context retrieval variable length very interesting, can you be more specific?

On Sun 14. 5. 2023 at 14:13, lookevink ***@***.***> wrote: hey all! great community. I'd just like to point out that to be an AI entrepreneur, you don't necessarily have to be one of these flashy startups with mind-blowing tech. you can commercialize one of the existing solutions and refine them. I'm currently onboarding our first two customers (both pretty big names) for internal/external context aware chatbots. I made the context retrieval variable length, so the intelligence is miles ahead of other ones. look for a need, and create/find the right tech for the solution — Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A6MDBXU6IENW6TSJYJ46AZ3XGDD7ZANCNFSM6AAAAAAWJ2EWDQ> . You are receiving this because you commented.Message ID: ***@***.*** com>

-- *Matěj Havlín* Phone: +420 733 154 483 E-mail: ***@***.*** Messanger: *https://telegram.org <https://telegram.org>*

…

----- Obsah této zprávy není návrhem na uzavření smlouvy ve smyslu §1732 Občanského zákoníku. This message contains confidential information and is intended only for the individual named. If you are not the named addressee you should not disseminate, distribute or copy this e-mail. Please notify the sender immediately by e-mail if you have received this e-mail by mistake and delete this e-mail from your system. E-mail transmission cannot be guaranteed to be secure or error-free as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or contain viruses. The sender therefore does not accept liability for any errors or omissions in the contents of this message, which arise as a result of e-mail transmission. If verification is required please request a hard-copy version.