We need a good solution for non-techies to install and operate dweb I think … it’s so simple and super powerful to build apps on top of it … but in this moment of time there is absolutely no chance my parents would manage to use it … (same issue with ant-tp … even the autonomi ant-cli …)
…I could test running dweb via tauri and connecting the friends frontend to it. Since with tauri I can execute rust before/when running the app this might even be pretty straightforward. Then there could be precompiled standalone-releases for non-techies that come bundled with dweb … (this just raises the question about the SECRET_KEY for dweb … ideally we would create a way to use an “invite” (just a few cents in ETH(arb) and ANT to cover the fees to initialize the app))
…let me think about it … I think that may really be possible by bootstrapping from others that are willing to cover the cost for the 2 needed scratchpads … which is absolutely neglectable …
I’ve an update to dweb instructions which goes some way and think we should be able to discard the environment variable, but haven’t tested that. I think it will pick up the local wallet now.
But it’s a stretch to think many can jump the crypto wall. IMO we don’t have a way to that yet and I don’t see it myself.
that’s why I was thinking about maybe sending some coin to a temporary wallet, that is the “invite” you can share - for the one accepting the invite the app then sends the coin further to a self generated key
the user would either need to be a crypto bro … ooooor know someone who is and can generate an invite …
but you’re right … we only get 1 hop further there … (but maybe that 1 hop is enough to close gaps …? - since invites could be forwarded too ..?)
… with friends essentially utilizing 2 scratchpads … there would be just the 1 time fee for setup and after that it should just run …
another point that is causing a bit more headaches now with the switch to WebRTC is how to test and evaluate the mesh-communication later on …
… since it’s all running client-side in the browser I’m thinking in the direction of spawning many DigitalOcean droplets with activated firewalls and launching multiple instances of Friends per machine via Puppeteer to connect them and see if (how performant) the mesh networking works and delivers all messages reliably to every participant (and at wich speed)
if someone has a better(/different) idea feel free to throw it just in here - I’m not super eager to spawn 20 droplets with 20 friends instances per droplet (automate the usage of friends via puppeteer and to collect the data at a central point again to be able to see and diagnose what happens) to test a 400 participant chat …
… though that’s the only way I currently see testing to be possible at larger scale xD
AI has very little role in code of this kind and I won’t accept AI generated code in PRs for my projects. It just isn’t ok for people to submit code they don’t understand themselves and ask me to review it. Fine if people want to do this in their own projects but I ask people not to waste my time with it on my projects.
In my own work asking an LLM is a valid part of the work, but a lot of the time it suggests code it is in error, so using it to generate code is not a good idea. It can help point you in the right direction, but often sends you the wrong way too. So I prioritise other resources first.
And even if it starts to generate working or even good code, I really don’t want code in my projects that I don’t understand.
Buuut.. IMHO AI is OK for grunt work such as generating and organising CSS and as @riddim says above, guiding as to which property to change for visual effects. It’s a cheat sheet on steroids. But all the cheat sheets in the world won’t make you a (much) better programmer.
oh well - I think AI is a very useful tool but as with every tool you need to use it the right way …
…if you don’t understand the generated code you most certainly didn’t ask the right question / you don’t know how to evaluate the produced output and if it really resolved the issue you wanted to address … and you certainly have no idea how to long-term-maintain it …
but letting it look for e.g. where the port for the started dweb app is being generated by @happybeing and then knowing where to tackle the issue without having to search all of dweb by hand would be a very valid use case imho xD
(oh well - you said yourself asking it and letting it suggest stuff is what you do too )
This is true but it’s a bit worse IMO and I’m backed up by a lot of people in the industry who actually write code and understand how to maintain it.
The problem is if you think the code is good, and it even works, but in reality if you didn’t figure out how to do this and didn’t write the code you are at a disadvantage right away. And these machines are not designed to write good code. They are designed to generate plausible looking output. And then to make it do what appears the right thing etc.
At the end of this, if it is code you want to maintain, or want someone else to accept and maintain, using AI generated code is not acceptable for me. One day maybe AI will do the whole job but I don’t see this happening until humans are out of the loop, and by that I mean we don’t have a say in what the AI is really doing.
Because the only way humans can truly specify exactly what they want is in code. You can’t do it in English, and if you use formal methods you are doing something even more rigorous than code. So the way we get apps that do what we want is to:
have an idea
think about it
maybe prototype it
design
code
get feedback
iterate
and so on, forever
Ultimately a non-skilled coder will be able to tell an AI what they want at each stage, but doing anything complicated like this, or mission criticial (even in a business sense) or understandable by humans, is going to take much longer than if you are talking to Engineers who occupy the same world as you and who can understand all the things you don’t know you need to tell them for you.
AI is a useful tool, it’s just not the tool people are being told it is when it comes to software.
…
, not convinced that she fully understands what she does on FB after all these years either.