Great sugggestion

yep, have a look at open interpreter which is designed to do just that.

maybe, it's the sort of task I would break down into steps - the 'find and install from vscode marketplace' would be the bit I think it would struggle with the most. Personally I would make a small script to do the hard part and then you can test it in isolation and use OI to integrate it with other shell commands if you want.

RTX 3060 so pretty light weight

@@LearnCodeWithJV I have the 4060TI, would that heavy one run well on mine?

thanks

yeah, lots has changed since the original. It's on my list.

can't wait!@@LearnCodeWithJV

yeah, it's a shame they couldn't make a go of it

Nice, the good news is the open models keep getting more performant!

Probably. I know google colab doesn't like running servers and prioritises interactive notebooks so I would guess it's a bit of a moving target. I've seen a bunch of stable diffusion notebooks they've disabled but the Ngrok + Litellm combo sounds simple enough that it would be worth a try.

My pleasure. It's great to see all the progress on the project, you're writing code a lot faster than I make videos :)

I get a few questions every now and then when I run local stuff but the setup isn't that interesting and I don't know if I'd have much more to say than list some specs. NVIDIA RTX 3060 with 12 GB is the main driver of how well the local models perform. For the rest I just run Windows Subsystem Linux on a i9 with 3.3ghz and 64GB memory

Thanks

NVIDIA RTX 3060 with 12 GB is the main driver of how well the local models perform. For the rest I just run Windows Subsystem Linux on a i9 with 3.3ghz and 64GB memory

Glad it was useful

Glad it was helpful

yeah, it's name is apt. I've seen a few derivative projects kicking around which are claiming significant speedups. I'm waiting until I find one with good voice cloning and multi lingual abilities and a commercially available license to do the follow up.

Thanks!

i9 3.3Ghz with RTX 3060 using wsl2

I remember it only showing up some of the time and being disabled when a voice was rendering. Have a half memory of it only working for built in voices but I can't recall.

indeed, people are building a lot of cool stuff.

I don't know if I could handle it as a wake word, but I love it in the command line.

@@LearnCodeWithJV an Ozzie intelligence.

which project were you referring to?

It looks like you don't need the " in front of the echo

I'm using oh my zsh with the bira theme and hte git, node, tmux, docker plugins - nothing custom.

Not yet, I created an issue that hasn't had any action on it - github.com/KillianLucas/open-interpreter/issues/214 - feel free to add your voice there if you'd like them to prioritise it. You can get access to a (as of now) version of main with the files removed at github.com/joshuavial/open-interpreter/tree/clean-git-history - you shold be able to rebase it off the official repo as needed, but as soon as you add their remote to your repo and pull it down you'll get the large history anyway.

@@LearnCodeWithJV Thanks for that branch. Yes I hope they resolve the 214 issue over having others wanting to git cherry pick.

Merged! Thanks to JV :) @@Care2WorldBuild

Thanks for the feedback, talking slower and clearer is something I'd like to work on.

yes, it is something to watch. I found that the local models were good enough for some simple tasks.

​ @LearnCodeWithJV I have created something very similar, been working on it for a while, and changing it to GPT 4 made it a lot better, but also alot more expensive. And for me its that i need the conversation to get the results back from every command it executes so it may stop its task or change the task objects. There is a huge potensial here when combing it with better prompting (tree of thoughts or graph) but that would make it even more expensive. Also connecting it up to a credetial db for access to other system that what it executes on. But trusting it to perform something in production is not there yet, it would have to be supervised by a human.

Thanks!

Yeah, it’s amazingly useful at miscellaneous admin tasks. Nice use cases.

My pleasure, hope you find it useful.

Maybe, I think there is a big difference between measuring bias and training models for certain outputs. I think it should be possible for quite a broad group of people come to agreement about what an accurate measure would look like, while I suspect there will be a lot of differing opinions about what the desired output should be.

@@LearnCodeWithJV The problem with bias is more fundamental than that. The measurement tool itself will have bias. No broad group can create an unbiased tool. Think of it like this, let’s say they create a tool and make it perfectly unbiased by today’s standards. Will people 500 years from now still think it is unbiased? No they will laugh at what we considered bias. 500 years is a huge divide, but we don’t have to go that far. Just take any two opposing groups of people today and you will get wildly differing opinions on what is bias. I work in Silicon Valley as a software engineer. My coworkers and management are extremely left biased. There is no way that won’t make it into the tool. Just asking the question about “female nurses representation” vs “50/50 representation” is already left biased, and we are just asking the question and haven’t even gotten to the question yet! And if a person can’t see that, then that is their left bias blinding them to their own internal biases.

Noted - you might want to checkout openrouter.ai - they are an easy way to get access to a whole bunch of unfiltered models without worrying about self hosting.

They are indeed. Thanks!

Have you played with cursor at all? It's still on my list to give it a kick - the way it handles docs looks to be the most interesting piece

@@LearnCodeWithJV I’ve been using it for a month. I found 2 things: 1. It worked a lot better with tabnine/copilot + vim. 2. I like some of the ui simplifications they made to vscode Regarding the chat functionality, so far I haven’t found it hugely useful. I’d ask it to explain stuff or bring up the docs for something. But it’s not as in your face valuable as tabnine/copilot

I wasn't aware of that - thanks for sharing, Updating the description now.

That's a fair point, there are a lot of companies that won't allow folks to use OpenAI apis. I expect that will be solved reasonably soon though as companies stand up enterprise friendly AIs or the open models get good enough with fine tuning. The main thing I was surprised with Aider was that the quality of the code + the dev workflow was good enough for production use.