Auto-GPT

I've installed and played around with Auto-GPT and here are my findings.

1. ChatGPT does not have a hyphen in its name but Auto-GPT does. How about that.

2. I thought I needed to install Pinecone before I could start using Auto-GPT but it turns out I didn't. (I think Pinecone is only needed if you're running very long tasks in Auto-GPT which uses a lot of memory.) However, I also saw this warning about Pinecone. I didn't fact-check this however but you might like to if you need to use it.

3. Here's a website that provides easy to follow instructions for installing Auto-GPT and all its dependencies.

Please note that:

a) If you're on a Mac, then you need to run python3 everywhere it says to run python. Otherwise, it installs fine on Macs as well as PCs, and Linux too I believe.

b) The instructions mention a .env file. There is not actually one provided but there is a .env.template file. Just rename that to .env.

4. When I ran Auto-GPT, I got the output shown in the image further below. For readability, I've shown the text that I typed as italic bolded and the key prompts from Auto-GPT as bolded.

5. What I noticed in running this is:

a) There is a hell of a lot of output, however, thankfully, the most important output is written to text files.

b) Auto-GPT assumed I was interested in homeless people in the Los Angeles area. I then asked it to consider only homeless people in the Adelaide, Australia area. It was fine with that, however, after a while, it reverted back to the Los Angeles area.

c) Sometimes it appeared to be going around in circles although it seemed to eventually stop doing that and continue to progress through the task.

d) It had a problem accessing Chrome at one stage (it Googles quite a bit) but this may have been because I had been running another instance of Auto-GPT in another terminal session.

e) Auto-GPT uses the OpenAI API to access the GPT-3.5 service. When I use this service through one of my apps, I'm flat out using more than a cent's worth of service time. However, using Auto-GPT, it's very easy to use 20 cents worth in a single session. It's obviously still very cheap but, as you can instruct Auto-GPT to run any number of steps at a time, a wrong, very large, number could end up costing quite a bit.

The obvious benefits of Auto-GPT over ChatGPT are that it can:

a) Write files to your disk.

b) Google to find the latest information, whereas ChatGPT knows very little after September 2021.

I also think the great thing about Auto-GPT is not really the fact that it may eventually reach an endpoint and achieve the goal(s) set (I didn't run it long enough to see what happens) but that it gives you quite a few good ideas on what you should be doing in order to achieve the goal(s) you've set it.

Hope this is useful.

Output from a run of AutoGPT