Ah, I thought you meant there was some technical issue that I wasn’t aware of / it had been superseded by a superior method.
Ethical issues aside, the arr stack is a good gateway drug for self hosting. It’s fiddly but probably in a useful way. JF/Radarr/Sonarr/Sabnzdb set up genuinely introduces a variety of branching skills.
Arr leads to self hosting, self hosting leads to home assistant, home assistant leads to … suffering :)
Your en dash is showing, Wesley.
“A lot” is doing some Olympic level, Gold medal performance lifting there lol.
Lemmy is loud in its AI hate, nuance be damned. I’ve seen drivebys on !LocalLLama (notably on a project about AI infinite radio).
Apparantly, everyone is now an expert on ML and AI (I sure hope they didn’t use Gboard or Apple to glide type their message or any sort of STT), just like everyone was an expert on epidemiology, politics, trans rights, ethics, pop culture etc 5 mins ago. You don’t need to be an expert to have a position but at the same time, empty cans make the most rattle.
What’s up with arr stack tho? I’ve been pulling direct from 1337 of late (and even more recently, just using CloudStream). Are we “no one is gay for molemen” on the arr stack?
That site appears to look for superficial signals (commit messages, claude.MD file). I wouldn’t put too much faith in it for detecting “slop”.
Ironically, the best way to see if something is slop (if a person was so inclined) is to point Claude or ChatGPT at a repo or code example and ask “did a human code this? Why or why not?”.
Unsteered Ai tends to produce immaculate code with logic gaps. Humans tend to produce weird shit that somehow works and actually makes sense when you step back and look at it.
Probably the best tell is when code clearly bears the marks of “oh, this guy got fucked by this problem; I can see the patch job. It’s not elegant but it makes sense if X occurred, which is what Y appears to be for.”
I believe “opinionated” and “belt and suspenders” is the Americanism for that. Kludge and bodge are also true.
IMHO and YMMV.
Genuinely disappointed that didn’t lead to a rick roll. Also, did you mean ZnVjayBvZmYgbWF0ZQ==
Irrespective, good work on the app. It’s not something I have personal use for but I salute the effort. You didn’t need to make your own personal project public, and you sure as shit didn’t need to post it to a hostile site (lemmy has a hate boner for AI something fierce).
The fact that you did anyway isn’t nothing.
Next time, write more of the intro post yourself :) “What I built” and “I’d love to know…” are too Claude shaped not to rustle jimmies here.
Step 1 (gateway drug): Jellyfin
Perfect. Then my third suggestion stands - get it make you a bespoke training module.
Create a project container: upload the source documents already cited by others / ones you find useful. Just 4 or 5.
Add the rule “You are helping me learn ____. You must not provide full refactors unless specifically asked for. Use Socratic method where appropriate”.
Pseudo code what you want to do (hell, draw a flow chart) and ask “based on what I want to do, your background knowledge and the contents of the container, what’s the first thing I need to learn? How should I approach this?”
Then ask “why?”
Hell, get it to make you coding exercises.
After each session, get it to make you a handover note in markdown format. Download it and then add that to your container. (Later on, you can get really fancy and start making your own llm-wiki)
People shit on LLMs as a knee jerk thing… but coding IS a language…and if you understand the logic, the rest is syntax. Also, it’s not as if you’re making some million dollar mission critical thing- you’re teaching yourself through successive approximation.
If you don’t feel comfortable using online models, there are even local alternatives.
I still think JavaScript is the faster way for this, but who knows.
Oh that’s cool! Cheers for that.
Scoping first, code ideas at bottom. Kewl?
Is this a “learn it by coding it” project or is it a “I want this thing to exist, no one has done it, but my code skills aren’t quite there” project?
If the latter, would you consider iterating via llm (as you mention n8n, so I figure you’re in that space anyway) or is this a purely a learn by doing thing?
Come to think of it, there is actually a third option here. You could get the LLM to teach you how to code it by writing some pseudo code and asking it for pointers / starting steps. Claude web is pretty good for that sort of thing, I think. You can get it to tailor its lessons to what you need without the tedium of starting at “Hello World”.
You seem like the sort that could keep that interaction honest and not let it just do everything.
PS: I read about your setup - sounds brilliant. Go you good thing.
PPS: n8n has a Code node (JavaScript), and parsing that weather JSON into a formatted string is probably like 15 lines of code. Something like -
const data = $input.first().json;
return [{ json: { temperature: data.list[0].main.temp } }];
add a Code node after your HTTP request in n8n. Get one single value out first. If you see a temperature number in the output, you win.
Pulling Python in just to parse JSON is probably adding a tool you don’t need for this.
JavaScript may be ass but it’s literally there, so it’s omnipresent ass. :)
Once you’ve got the JSON parsed, turn it into one small HTML weather card.
const data = $input.first().json; const item = data.list[0];
const html = <div> <h3>Weather</h3> <p>Current temperature: ${item.main.temp} °C</p> <p>Feels like: ${item.main.feels_like} °C</p> <p>Humidity: ${item.main.humidity}%</p> <p>Condition: ${item.weather[0].description}</p> </div>;
return [{ json: { html } }];
(Sorry about the formatting ; Lemmy formatting is weird AF)
Anyway, as one journeyman to another, that’s where I’d start poking. ICBW.
Let’s tag it as “provisional” then. As in, once you have my provisional trust, accrued over time, I’ll probably stop auditing every single line. I’ll still look tho.
But the long and short of it is this - XZ utils backdoor actually makes case for trusting clankers more than human collaborators. Clankers are incompetent… they usually aren’t Machiavellian.
I’ve heard it said that an LLM is like a Labrador retriever when it comes to coding. Overly excited, pulls ahead, does some really goofy shit and sometimes chews up your couch (hello Qwen 27B)…but it is trainable.
Human devs are like cats…which is oddly on brand for this project :)
I’d sooner trust a clanker I had prompted with my house style ticket and narrowly sandboxed than a rando online. Of course, the difference is, a rando may eventually earn trust…a clanker doesn’t - but it doesn’t need to if narrowly scoped.
EDIT: here’s a template I use / created for Qwen / Codex. It’s…opinionated and bears scars of prior over eager Labradors. This is usually step 1 I fill out. My fingers are going to shit with O/A , so am trying to minimise scut work.
Status: PROPOSED Timestamp: DD-MM-YY-HH-MM Priority: P0 | P1 | P2 | P3
One paragraph:
Describe:
Include: We do not want … We do want …
Hard gate.
Before implementation exists, prove:
If proof fails: stop and escalate. Do not patch.
Each gate:
Mix of:
Prefer: prove behaviour changed, not just coverage increased.
Minimum acceptable state.
Must describe:
Stretch target.
Usually:
State assumptions explicitly.
Examples:
Describe:
Prefer: small typed objects.
If adding logic:
prefer:
Avoid:
Policy: config
Signal: code
Config controls behaviour. Signal detects reality.
Explicitly include:
Explicitly exclude:
Numbered list.
Must be testable.
Agreement on:
Only then may implementation tickets follow.
C’mon meow…
That’s kinda shallow, dude. But you do you.
It’s not like it’s named XxxLovesToSplooge69Xxx or LoliChanFingerBot.
Ironically, with a bit of effort, it’s quite possible to get an LLM to mimic your own house style and replicate it quite well.
You can use it as an editor or hostile reviewer or just bounce ideas off it, as long as you own the process.
Language transformation is literally what they exist for.
But the fact that the article failed to do this and produced generic slop is the actual insult. Not the fact that they used AI.
Like, we get it - portfolio slop / LinkedIn brain / look at me, look at me.
I don’t mind you using AI, but could you at least steer it a bit and do more than minimal effort? Or should I get my ai to talk to your ai?
True…but the arguably better / more defensive stance is “accept no PR unless the user explains wtf it does and/or I personally trust them”.
Iow, stop accepting PRs from randos - clanker or meatbag - full stop. The lowest cognitive load is “none”.
I don’t know you / we can’t have a convo why you sent me this? Into the bin.
(In my humble opinion, for a small or new project, that’s a cleaner footing anyway)
The claude.md file is cute, but I don’t think a claude would actually be tripped up by that.
It’s not such a high bar to pass to be honest with you. You’d probably need something more subtle, at which point you’re just shooting yourself in the foot.
The meow thing is more like a philosophical line in the sand than anything else and I respect it.
But given the way that Codeberg actually blocks crawlers and agents (and how Claude works), it probably doesn’t really do what we think it does.
No reason not to… except people tend to have bad reactions when a repo contains CLAUDE.md, what with anti ai sentiment being what it is.
In this instance, someone (correctly) read the file first and found the hilarious SuperTrooper-esque poison pill.
Disagree with the phrasing myself.
Yes “please ensure …” is nominally more proactive but “be prepared to explain” should force a technical framing in the first place, because the poster in on notice they may be challenged. Don’t write slop, read the room etc. It also explains why the rule exists / what the fail state is.
The first one just sounds like so much corpo HR double speak.
Be bold. Say what you mean - “We talk shop here. Be prepared to put up or shut up, else bonk”.
If the user can justify their question (if/when queried), it doesn’t matter whether the question is hardware, software or other - it will have a grain.
Additionally it might give people an foothold into OPs thinking process / whether they selected the right room in the first place.
My $0.02 and YMMV
I lol’ed (lolcatted?) but isn’t the better solution not to accept PRs from unknown / untrusted sources - ai or human?
Additionally, Codeberg is actively hostile to crawlers and ai agents isn’t it?
Still, this is funny. Not sure Claude would fall for it, but funny anyway.
Can’t be ChatGPT. Not enough goblins. It’s obsessed with goblins of late.
Ditto here. Rule 3 was…unevenly applied.
Pain. Suffering. Death.
(I’m kidding…or am I?).
I always think of home assistant like that scene in V for Vendetta, where he set up an elaborate circle of dominoes, and everything is working perfectly right up until he gets to the last domino.