A little bit tangential but given that I am genz. I wish to point out the fact that wordings changed from super deceased because I feel like the worth "death" is censored in social media so to prevent that censorship.
Corporations literally (sort of) decide how & what we speak xD
HAND (Have a nice day!) [just thought of this acronym because I love saying have a nice day haha]
👋
Haha I am genz 17 and I don't understand the gen-z slang either.
The feedback to write long is genuinely lost in all aspects and can make one feel alientated in my youth.
I had to learn gen-z slang somewhat too so if you need my help just message me haha and I will see if I can help ya!
Doesn't japan have metros although I have heard them be described as pretty congested.
I really love metros in my own country too and they are usually less crowded and if one can sit, actually really nice place to enjoy/even mobile phone internet access is pretty good 99% of times
wishing to learn more from you though as I feel like this is something that I can learn and has definitely caught my attention from many days so I wish to learn this art haha and I hope that u can teach me more about it
Any particular resources or anything would be helpful too!
Thanks in advance!
not sure if its the model faults.
I don't know, Sorry for pinging you or messaging you too much about it haha though but I guess I have been unable to replicate what you had done with rust/oaoh after trying out individually/even learning from what you said to me previously so yeah :<
Because I really wish to do so because this has really captured my eyes for many days but I am unable to really oaoh a project even after trying it.
Usually it feels as any model starts hallucinating even if I am holding its hands. I feel so especially in any visual context
I think my point is that oaoh really describes how I feel like AI (should?) be used, not the cursed cursor method which burned 5 million$
@emsh.cat Anyways on serious note, would there be any key idea/(manifesto?) behind oaoh.
So what would we say this
I oaoh'd a project?
I mean (phonetically?) this sounds like oyetched or oyeyetched
I mean am I wishing to create a word now called oyetch or oyeyetch?
To be honest I am just speaking it out of fun nothing too serious lol
So its much more of a joke but still haha
in a similar fashion to how emsh did. Because quite frankly, he has kind of shown a large company like cursor with their burning of 5 million $ budget who wanted to create an alternative (of browsers) that he (a human) with agent in loop can write a better alternative which he ended up suceeding in.
I used a new term to mention to something that emsh did (considering I am genz and my generation loves making slangs)
oaoh = one agent one human
Basically meaning that when I say for someone to oaoh a alternative of X
I mean to suggest that person to use LLM assistance to write alternative.
Here's one about repurposing the @lichess.org game compression algorithm as a chess engine and using it to analyze game 1 of the recent world championships.
follow your commit order
then to second this (sort of?) working prototype
Although. I guess I would only be able to use it after I get to one of the more last patches as Mac support was added in your project in the end and I suggested the AI to do it but it said that it preferred option 1 which was to follow your commit
Also it started to write python code to see the lzma compression or something in png file and I was seeing it but then I was like enough is enough and I rejected what it wanted to do and said that this is the issue and I did it twice
First from completely black -> black individual boxes
Because beforehand this was showing as black text and I had shown previous images too in github when I had one shot it was the same issue.
Then I pushed on the issue in this new modular thing I am experimetning and first the only thing it did was split the black box from one large to many small
This probably might be the reason why your project in the first place was successful too over the 5 million $ idea
We need humans to spot the blind spots & as such be in the loop. Completely autonomous agents would simply compound the errors caused by those blind spots.
Going to now have it run from commit 4 and so on and after some commits of major (UI?) development, I will run it and check it and intervene.
I feel like I got more progress than having it generate from scratch because I spotted a "blind spot" of it which was causing issues and insisted on it
So I ended up having kimi build a very mini tool in the first place
gist.github.com/SerJaimeLann...
Like you mentioned divide and conquer. I thought that you already *did* divide and conquer, why not just use that
anyways the main.go code for just converting github project into some info.txt & a list of commits/diff patches (I dont know if there's standard tool for that)
But my linux pc kind of got broken :< so I let it run till 3 commits and then I had it create a commit of screenshot ability and added it to mac and then saw it having issues (black boxes) and then asked it to fix it and in two attempts it atleast showed me something!
@emsh.cat
I did something which I wanted to share with you.
I created a go code which could then have all of your git commits and everything with git patches & I then asked Kimi's code to incrementally do things
I had to things a bit different because I am on mac (although I love linux)
Let's try the divide and conquer approach which you mention. Hopefully it works
I think an issue stems out of (gui?)
So in your first commit, from what I can tell you were doing x11 window with basic hello world right (Sorry I don't know rust) but I see x11 mentioned in your first commit
Ooh alright thanks
I assumed that given the code in the rust project was very less (20k) I thought that kimi could one shot a golang port.
Looks like I was wrong but then again its still a good experiment
I now get the feeling that agents cannot be asked to completely one shot large projects
@emsh.cat
Looks like Kimi K2.5 can't really create a gui for mac, I mean it can but it always end up being really glitchy really nothing in comparison to your project.
I don't know if its something wrong I am doing or that the model itself is limited
Thoughts?
Woke up and asked Kimi to fix again, I mean its better than the last so a little bit of progress perhaps? but still pretty bad fwiw
@emsh.cat
www.youtube.com/watch?v=L3wK...
xD, I had lots of fun seeing your project & playing wiht the go code too lol, gonna play it after I wake up too. I think the code is on github but I Haven't changed how much ~lines of code it is but you can check the code if interested
Lol you just responded to my HN comment as well
github.com/SerJaimeLann...
Anyways you can read my edit/edit 2 on simon's comment.
This is what got of a direct non human interefered golang port, maybe kimi isn't that good on real life use cases or we'll see.
Hey update looks like my project did take all the kimi credits i had for the go project.
I mean I wouldn't call it anything good. Its really glitchy but somehow it does show something on simon's blog post (even if its extremely glitchy rotated image or something with black lines)
