@Lensflare People believe in Jesus and Climate Change. If you have strong beliefs about one or the other and believe the other side doesn't have a sound basis, you've understood nothing of what I just said.

I fucking love Flat Earthers. I'm a Globe-head myself, but I have mad respect for people who can hold that belief.

The Rhine? Are we talking about Cincinnati .. or .. Germany?

The funny thing about Plato's Cave, or the Red/Blue pilled Matrix, or Christian Gnosticism .. or any of the 100s of similar duality/liberation stories since the dawn of man:

You have no idea if you're in or out. Imaging someone went into the cave, saw the "reality" and it was amazing and beautiful and then went out to the "real world" and said they were living a lie. The reality of our world is inside this cave! And then they killed that man.

Was Jesus the savior or a lie? Was Abraham Lincoln a hero or a tyrant? Was Caesar a great Emperor or the failure of Rome.

It never matters what's true. It only matters who wins. They get to write morality. They write the truth. Was in "actual" doesn't really matter in a collective society. It's literally an unknowable.

What models do you plan on running? I've found all the local coding models that can run on a 3080-Ti to be pretty terrible. Are the larger local models better than Claude4 or GPT?

I find the chatbots utterly annoying and worthless. I hate how their bullshit is now crammed into search results on DDG/Google.

What would you use local LLMs for?

You can point a dependency in your virtual environment pyproject.yaml to a local project. You can also reference a local project in VisualStudio instead of the nuget version....

..pretty sure you can do this in every package manager out there.

@jestdotty Me too. It's been in a slow decline for a while:

https://battlepenguin.com/tech/...

No, that's literally me. I got my masters in 2009, and my bachelors in 04! 🤣

#iOld

I use QOwnNotes. It saves stuff to a little git repo it maintains for each update, so you can brows your history using tig.

The trouble is LLMs are really garbage. You do need to have good engineering experience to critically analyze the stuff it gives back to you. I'm always more shocked when it actually works .. and I'm always refactoring it anyway to make it not shit.

I can't stand when people swear by AI and their new God. I know one dev who is very "bullish" on LLM coding, but he also skims all the code and doesn't really analyze it. It's just going to introduce a lot of technical debt that's very difficult to debug if you're not careful.

He told me he wouldn't hire a junior engineer today, and that kinda puts everyone in a rough spot right now.

Comments on posts don't seem to work on Librewolf (Firefox).

Upvoting a post takes me all the way back up to the top.

Let a rant/note. Looks slick. I'll add it to my daily. Good luck!

Edit: oh and add a favicon. Otherwise you get lost in the sea of tabs.

@12bitfloat same. There aren't even a lot of people on here for it to have been that slow for so long .. some admin probably banned some crawlers or some crawlers finally slurped up all of devrant.

@AlgoRythm I'd take Linux 100% anytime over Windows. I know Linux bottom to top. I built a Linux From Scratch back in 2001 and used it for two semesters and have spent most of my life on Gentoo.

If there's a problem in Linux, I can ALWAYS fix it because I can dig down however far I need to in the code. At my last shop, we had a really slick Nix setup to unify dependencies for all Mac and Linux devs. Even in Java, the built is always sane with Gradle, SBT or hell even Maven (🤢)

I did a whole talk on Docker/Nix and the history of dependencies: https://battlepenguin.video/w/...

Windows is just dogshit all over the place with horribly broken tooling, and no amount of .NET Core or containers will ever fix it.

...also my problem seems to be a know issue and I was directed to another project with an entirely different more convoluted way of pull stuff from Mongo to avoid the issue 😩

Is ColdFusion still maintained? Sounds like ancient technical debt!

Almost every first generation:

Now remove the useless comments

Now rename variables to these conventions...

Now .. Just stop. End execution. I'll fucking do it myself.

I don't understand people who just use prompts and never write anything by hand now. Someone told me they've made it through 3 months of code reviews without writing code. I'm like, "Is your development team incompetent morons who don't give a shit?"

I have to refactor nearly everything an LLM gives me so it doesn't look like garbage .. if it even works the third time.

Why bother storing the password at all? Just accept anything .. but first, generate a random number and check if n % 5 == true. That way it will fail sometimes. You should make it look like you're doing something, while also confusing people with password managers.

Why bother storing the password at all? Just accept anything .. but first, generate a random number and check if n % 5 == true. That way it will fail sometimes. You should make it look like you're doing something, while also confusing people with password managers.

Discord .. that's the chat app with all the potatophiles, right?

I once worked a job under ITAR regulations. It was pretty boring stuff honestly, but I'd have to verify someone is a US citizen before I could talk to them about it.

You can only deal with so much and it's all relative. Once upon a time, less than 100 years ago, you may not have know about the war going on 5,000 miles away that wiped out a village, or the British creating a famine to wipe out the Irish.

You really can't concern yourself with the kid drowning in a lake in Romania. Your local group is literally the most important for any practical purpose. The network of propaganda has led you to think you should care about arbitrary people in random parts over the world than your own.

It's impractical which is why the Jesus story says, "give everything your own to the poor." It's the unachievable ideal. When you chase it, you tend to develop "luxury beliefs," ideas that sound great on paper but ultimately hurt those you think they will help .. like "defund the police" or "redistribute wealth" ... one which leaves everyone dead and the later which leaves everyone starving.

If any of your are on the fediverse, I'm @djsumdog@djsumdog.com. (It's my Pleroma server, but you can follow it from Mastodon .. if your server doesn't block mine(

I use the fediverse more than anything else these days, and even then I don't use it that much.

@whimsical but it's not just me right? Performance is pretty terrible, considering how few users are actually on this site now.

It's just call the weight of living. You'll eventually get out of your rut.

I once sold everything, quit my job and owned only two backpack for 11 months, living out of hostels:

https://battlepenguin.com/philosoph...

Years later I did 5 months living out of my car driving across the country. I think I got all of that out of my system though.

I think it helped give me perspective.

@kiki Are you confessing through projecting?

@whimsical I don't use any Weighted Random Word Generators for my posts. If there's a meme or screenshot in my folder for the event, I look up and find the original post or article. When you see 2+ sources, one is usually where I got the idea and the other is the tracked back original source. A lot of times I find misleading articles or videos referencing something that says something entirely different.

LLMs might be okay for some coding (only in small amounts and you have to watch them carefully) but it's really dumb to use them to summarize news. They can't lie; they have no intent. It's random weights and an associated RAG store/vector search, etc.

James Corbett (corbettreport.com) commented on the endless e-mails he gets saying "Well ChatGPT says..."

I suggest people do their own research. When you do and write your own posts, you learn so much. My views have changed as I've written articles and done critical deep dives into stuff.

@D-4got10-01 FF9 is great. I played it entirely in ePSXe (PS1 emulator) way back in the day on a Dell laptop with a GeForce 4Go in it. Loved it. It's my second favorite of the one's I've finished behind FF7 (original).

I was at a mortgage company and after three years, I was the most senior person on the devops team. There were a lot of new people and the entire old team had attritioned.

A junior dev got promoted to manager after the last one got fired for job desertion. None of us were even allowed to apply for the position; they hired a director externally and promoted the junior. He did deserve it (he found like $100k/month in AWS waste) but many of us hadn't had pay increases in three years either.

@donkulator I dunno... seems like a great idea. Fool and his money and all that shit.

Was it down? I haven't been on here in a while since I got laid off, but I suspect I'll be back on soon since I got a jerb.

If you don't know what "embedding spaces" and "attention blocks" are, you probably shouldn't be using LLMs for code.

But I will admit, you are going to be left behind if you don't use it. But it needs to be limited in small block and it ALWAYS requires refactoring.

If you know someone who says "I haven't written code in three months and they all get through code review," that is a shitty dev team that's just racking up bugs and technical debt.

A cube would mean everyone got an equal payout, which is communism, which doesn't every work.

A sphere would mean people at the top and the bottom got nothing, while people who bought in at the middle made off with all the wealth. Who the hell would want to start a scam like that?

I'm glad you're talking about the context window and the model. So many people use these things as black boxes and have no idea how they work.

They can't "hallucinate" or "lie" .. they literally have no intent. What you have discovered is the truth: They are random next word (really token) generators. Anything they get right is literally by accident. People seem fine with things being right accidentally 80% of the time ... just waiting for that structural engineer who uses AI and it gets load bearing wrong and 8 people die in a bridge collapse.

3Blue1Brown has some excellent videos of how LLMs work, including the vector space addition, the transformers, preceptron blocks and attention blocks. There's a good 7 min layman's video too. I'd recommend going through them. They're eye opening.