OpenDevin

wait until GitHub Autopilot, indeed

Just tested Devin, myself! Honestly I find him not that impressive! However, I haven't tried GitHub's Copilot, yet! Thanks, Matt !❤

Neither 🙃

Devin

Installer with a few clics inside pinokio.

OpenVoice tutorial, hell yea! Take an audio book of your favorite reader and clone it.

xtts did that already

@@240dbprisms5 There is no working voice cloning in Finnish language yet, many others though.

yes he never used the github chat otherwise he would be differenciate because i think github copilot chat is the best even can beat the gpt-4 easily

kzfaq.info/get/bejne/fbJ7oqecx7mVc5s.html

How slow would that be on Pi? Just cos you can?

@@joythought I don't think the RPi are slow anymore, but arm64 architecture is getting common, was the point I tried to make.

Exactly!

what are you trying to say?

@@Marc-hn4eg that if you tell it not to lie, it won't

@@JustinArut it certainly is the type of prompt that could get preemptively answered... but didn't :D

I've also found the most common answer since using that prompt is "I don't know" - maybe the prompt needs work for the purpose of theorising

@@six1free If you ask it what does IDK mean? It also responds I don't know

I think we're already there. It's just an iterative process of going back and forth with the LLM to get what you want coded. The upside is now you don't need a whole dev team to make a site like kickstarter, facebook, twitter, etc. But if you want to invent something new... like when bitcoin first came out... you're going to have to get your hands dirty.

Carpenters and painters still exist

Soon enough…

I wrote machine code on z80 processors. Why would we want to go back beyond having the capacity to dig in if really required or for kicks?

I think that's going to be the case for a lot of A.I. use, especially as they become more useful in a lot more areas, as I can't imagine many people and companies wanting data going back and forth to the online server, as it's easy to see how that can be a security and privacy risk as we have no idea what they are doing with that data. Locally run models are likely the future as they are getting good enough and will continue to get better, without having any of the security and privacy issues, and they also have the advantage of not being as restrictive and can be fine-tuned for your given tasks. Anyway, if this was just a year ago, I would likely say it was unrealistic for locally run models to compete with the online ones, but with how quickly A.I. is developing, especially open source small models, it's becoming less of an issue that for a lot of cases, we can use a locally run model.

The whole idea is to get you locked into a software ecosystem. They're also doing this with cars now. Next, companies will be pressuring you, the programmer, into one or another because the CEO has stocks in one of them.

@@WhyteHorse2023 this. local self-hosted open source models or nothing. I wont be a API rent slave.

Top companies around the world are pushing employees back into the office. And yes I am in Australia.

maybe we need something like Github for HI. Like a way to trace belief systems since the beginning of thought and critical thinking. People review ideas like code..facts become verified. Kind of like X community reviews but with more structure..anyone feel like coding this up? I might have a go at it myself with all these new high level coders..

So many people said that about the Rabbit. But then the hypefollowers started foaming "you do not understand, it is LAM! That is ACTION model". To which most people just rolled eyes I guess.

"this" was referencing as-is, but what some people did was change it and adapt it to work. It's now "that", something based on r1's work adapted generic android ecosystem likely without the hardware specifics of r1. porting shouldn't have been too hard.

You mean some dude on KZfaq that didn’t know what he was talking about? Maybe check out Marvin Praison who actually got access to it.

Why is Devin so talked about when is not released yet?

@@neociber24 rage bait

Got debunked? How?

​@@jld-ni3vfIts kinda true though, it could do anything or nothing. Unless we can do thorough testing it does mean much.

@@JustinArut And yet you are here.

He actually just kind of reiterates the same stuff over and over again to stay relevant and the hype cycle. The funny thing about all this is how they’re selling it as this new digital God but the reality is, it’s just gonna be used against you to digitally enslave you. because in the end, it’s it’s control and people have a history of being sucky

100%but once you’ve thought through the solution why code it all when you can ask for it with everything defined and just edit the code

“unmaintainable crap” Old school thinking if in future no human will ever again be in the position of maintaining code.

If you’re a coder you’ll know that when you get into “the flow”, you are thinking and coding “as one”, you make decisions that will help you down the line, you known which parts of the code are going to be likely changed a lot, so you take care to make those parts easy to change. If you know, you know. My experience of coding assistants is they are only good for laying down boilerplate code - then they just get in the way with incorrect or bad suggestions that have to be read and disregarded - all spoiling ‘the flow’ I mentioned.

@@tollington9414 now imagine compartmentalising all that in a modular system you can then upgrade with a prompt. I am a coder and ive gone from maybe 100,000 a year, to over 10million. my logic is still there I still feel the flow now, when walking too. ill chat to an AI about the problem walking doggo. get a spark send it to the system. when im home its waiting for me to demo! how is that not better? I still program myself but feels like wasted effort these days personally.

This really sums it up … good for boiler code … bad for anything it hasn’t seen in the training set kzfaq.info/get/bejne/mqqFdpB91rzWgIk.htmlsi=WeIqQPfAa-fm9J6U

I'm not sure a better model, will implicitly mean "close to agi". Officially it's just a better model.

Or it's official GPT-4 is nowhere near AGI, which... yeah I'm gonna go with that one.