NateEag

The Web is not being developed "in collaboration with the machines."

Humans downloaded the Web and shoved it into the hopper of a sausage grinder.

When another human turns the crank, they may get something that can be made a part of the Web.

Are LLMs amazing? Yes.

Are we collaborating with them? No.

When we use an LLM's output to expand the Web, we are taking the hard work of those humans who toiled to build the Web and pretending it's ours, because when it's mashed into sausage, it's hard to recognize the ingredients.

Yes, some humans have built motors that can be used to turn the crank without direct human intervention.

The crank still turns only when a human starts the motor running.

I wish I had that kind of effortless motivation.

It is definitely not effortless - it's a battle. https://en.wikipedia.org/wiki/The_War_of_Art_(book)

Over the decades, it's become one I love, and I can see how the struggle has helped me become a better musician, writer, programmer, and person.

Being compelled to use genAI by my employers has largely removed that struggle from my day-to-day, and I hate it passionately.

The replacement of careful thought and fighting to create something compact, elegant, and robust with reading thousands of lines of mediocre generated code, slogging through page after page of uninspired just-good-enough dreck in order to catch the LLM's inevitable moments of absolute failure is mind-deadening and tedious.

I am figuring out what other careers I can swap to if this becomes the new normal.

I've been (grudgingly) using Claude since February.

When I started, I discovered to my shock that it was by far the worst TUI I had ever touched. Rendering glitches, keyboard input screwups, and just all-around jankiness.

Despite that awful baseline, I have to agree that it's gotten noticably worse in the past week or so.

It's started mangling my terminal sessions somehow, so that not all characters I input are visible in the UI.

Backgrounding it, doing a 'reset', then re-foregrounding seems to fix it.

Of course when I asked it to diagnose the problem, it assured me that Claude has no such big and it must be something else I'm running.

All I'm running is tmux and Claude, though, and the behavior surfaced while I was running Claude.

So, yeah - if they just recently updated Claude to use the Rust rewrite, this increased crappiness might actually be due to that. It does seem to at least roughly correlate.

And I couldn’t have ever “stack overflowed” my way to diving into a project in a brand new language.

It's a very tiny project, but I dived straight into a new project in a new language completely by hand this month, precisely because I'm miserable and exhausted from my job requiring me to drive a bullshit machine instead of solving problems by thinking.

Getting a few minutes in the evening to actually program, and learn new things as I do it, has been crucial to maintaining any sanity at all in this brave new world of unending heaps of slop.

Yes, they are.

They may be doing a poor job of it - time will tell.

But, at present, they're still the ones merging PRs and deploying changes to production.

As far as I can tell, yes, the workflow is "assume the LLM can fix it."

I've certainly seen people on this site defend that idea vociferously.

I won't tell you it's impossible - Claude does do legitimately amazing things.

It does not, however, seem to me to actually have deep understanding of what it's doing. "Amazing" does not mean "admirable" or "trustworthy."

Whether or not he's serious, as someone who's grudgingly using Claude at work due to mandates, the productivity gains do shrink massively (and sometimes go negative) if you actually attempt to gain the level of understanding you'd have of the system had you written it yourself.

Does it matter if the developers understand the system they maintain?

I guess that depends whether the genAI maximalists turn out to be right.

The documentation is truly a thing of beauty.

As someone who learned a lot from Non-Violent Communication, discovering mergiraf went something like

  > ...oh, cool - I've wanted this for years, great that someone buckled down and wrote it.
  >
  > Why 'mergiraf,' though? 
  >
  > [Sees NVC link on the site] 
  >
  > ...OH. Conflict resolution through understanding. That's absolutely brilliant, at so many levels.

Don't forget mergiraf, for tree-sitter-based conflict resolution:

https://mergiraf.org/

I once root-caused a revenue-losing bug that had the entire rest of my team stumped with ten minutes of git-bisect.

It only worked that well because the team had all internalized my advice to write small, coherent commits, so the bisect landed on a ten-line change.

I strongly dislike the recent trend towards squashing every branch into a single monster commit.

If people live lives wherein they build important societal systems to be biased towards "short fast exciting" lives, the opportunity to live a long, boring, safe one may just evaporate.

Yes, but whatever the LLM has authenticated access to, an attacker can convince it to mess with on their behalf.

It's maybe closer to putting a sign saying "The door is locked" on an unlocked bank vault.

It does nothing to improve security, and if someone manages to get inside and see the sign (i.e. "extract the prompt"), it gives them a strong hint there's interesting stuff behind this door.

It would be hard, yeah.

It's a network effect thing - if enough people used an OSM client and opted into sharing anonymized location data, you might be able to get somewhere with it.

In practice, I think it'd have the same problem as the core map data - surprisingly good data in places with a lot of adopters, and nothing at all in places without.

That said, though, the traffic data is pretty secondary - a good, open routing engine would be lovely to have, and will slowly become more relevant as the map expand and fills out.

It's not true.

At least, not exactly.

At its best, the OSM data puts Google to _shame_. Super-fine resolution and details, nuances of street layout precisely correct, mountains of meticulously-maintained metadata that makes manifold niche geospatial apps possible.

Alas, in any given area, OSM is only as good as the union of the publicly-provided street data and the obsessiveness of the area's mappers.

Most towns in central PA, for example, have no street address listed for hundreds of buildings.

I've made many small contributions to help, and really enjoy doing so, but someone looking for driving directions needs to be able to whip out their phone and slap in an address / business name and have it Just Work.

You can't reliably count on OSM for that.

I wind up falling back from OsmAnd to Gmaps probably every fifth drive.

I think our (assuming ur American) is the strongest its ever been and I didn't even vote for the guy.

Um.

What?

As a conservative-leaning registered Republican, I think Trump has brought the US the closest it's been to self-destruction since the Civil War.

His administration has authorized heinous human rights violations repeatedly, and has prevented "law enforcement" agents who killed innocent civilians from being brought to trial multiple times:

https://en.wikipedia.org/wiki/Killing_of_Ren%C3%A9e_Good

https://en.wikipedia.org/wiki/Killing_of_Alex_Pretti

The administration has violently pushed away most of its historical allies, and is going to enjoy the privilege of trying to stumble through the world on its own for quite a while.

The administration got the US embroiled in an utterly stupid, optional war that was guaranteed to have the harmful results it has: https://acoup.blog/2026/03/25/miscellanea-the-war-in-iran/

The economy looks good on paper, but that's almost entirely due to the current genAI bubble, not any intelligent economic choices by the president's office.

Most recently, the utterly idiotic ruination of the reflecting pool in DC, and subsequent insane claims that it was intentionally destroyed by the administration's critics is emblematic of the stupidity and self-destruction inflicted on the nation by President Trump. It's a small thing, compared to many of the issues, but it illustrates the harmful behavioral patterns in a crystal-clear manner:

https://en.wikipedia.org/wiki/Lincoln_Memorial_Reflecting_Po...

This take is so weird I suspect author somehow financially benefits from tokenmaxxing.

They might own a chunk of NVDA.

Thank you for your comment.

I had assumed that the recent OpenAI solution to an Erdos problem represented original mathematical thought.

I went looking for more details, and found this Scientific American article which provided some nuance I had previously missed, namely that the mathematicians involved don't think genAI created any really new mathematics - just applied what existed in an intelligent, elegant way:

https://archive.is/oNCSO

So, impressive though the result is, perhaps your claim has more grounds than I had thought.

Several thousand employees means several thousand chances per working day to create a security breach.

I suspect that smaller teams are, on average, more likely than larger ones to write secure software.

As a fluent native speaker who has read thousands of books and sometimes reads dictionary entries for fun, a number of these definitions are actually slightly off.

"Verbose," for instance, is defined as "Using more words than are needed."

That's not exactly wrong, but it's kind of misleading. "Verbose" explicitly means using a large pile of words, drowning the reader in far more words than are strictly necessary.

"More words than are needed" could be as limited as "used a three-word construction in a sentence where it could have been one."

There are many more like this.

Please, I beg all of you - don't use LLMs to generate linguistic slop that claims to be linguistic education.

I weep for the world that is to come.

My position is that inflicting horrible outcomes on a small percentage of the population is not justifiable.

If you don't understand your software well enough to prevent and/or fix such outcomes, you are in moral default by making it available to people.

I find it analogous to selling people hard drugs.

Some users will be fine.

That does nothing for the ones it utterly destroys.

The only way people will get to self-expression that's worth hearing is by working their way through the self-expression that wasn't.

In a better world that would happen in school and as a normal part of growing up, but at least in the US, that mostly doesn't happen.

If humans use LLMs to write, it's much more likely they will never uncover the unique perspectives they could share with the world.

So, for those whose parents and teachers failed them, embracing, encouraging, and engaging their clumsy, self-centered rudeness is, I think, the best path forward.

Granted, when it is clumsy, self-centered, and rude, one of the more helpful things we can do is offer the critiques that clearly and coherently point that out, as you've done here.

...though I stand by my claim that using LLMs to write is actually deeply offensive and rude, a rejection of what it means to be human, and that it should be met with strong rejections.

I agree that it was a rude, tasteless metaphor.

Alas, for rude, tasteless behavior, such as replacing your own authentic self-expression with the mellifluous spew of verbal diarrhea that bullshit machines slather across all surfaces they touch, rude, tasteless metaphors are the only fitting ones.

What's not okay is a world where unreliable tools can destroy people's lives based on entirely false information, and the purveyors of those tools and false claims get away scot-free afterwards.

If this ruling forces companies to put more money into #3, whereas now they're coasting on good enough, I'd say it was speeding up innovation.

The thing is, no one has the slightest idea how to stop hallucinations.

The models are fundamentally "hallucinatory" at core - they generate what is _probable to follow the string thus far in its training corpus_, modulo RLHF and friends.

Notice that nothing there has any rigorous relationship to truth.

Sure, the companies could start pumping money into pure research on what models other than transformers might yield something that can reason rigorously, but at that point you're talking about finding a way to throw out LLMs entirely in favor of a less-pathologically-broken model, like Gary Marcus keeps complaining people should be doing.

I could not agree more that Claude itself is a janky, hacky, crappy piece of software.

When management at $DAYJOB brought the hammer down and said, "Everyone has to use genAI all the time, OR ELSE," I expected to be blown away by the tool I was avoiding due to ethical concerns, aesthetic objections, humanism, and long-term thinking.

I was born away, but not in a good way.

The CLI is _bad_. I've seen it randomly fail to render anything at all on the terminal multiple times. It has a vim-mode, but it's painfully buggy, and I can literally outrun it - if I try to type too quickly after hitting Esc for normal mode, it just doesn't return to normal mode. It's I was keeping track of the bugs in the Claude TUI, but gave up because it was taking _too much of my time_ to do so.

If nothing else, I'd say Claude shows convincingly that success is not the default for vibecoding.

Yes, it technically does the job, and no, I don't think I've ever used a worse TUI.

I have not.

I've considered keyboard-driven WMs before, but have thus far gotten by using regular WMs with a few custom keybinds for window sizing, movement, and monitor placement.

As long as you have an automated algorithm for the basic layout, it works pretty well.

I still have to contendb with apps that don't offer the shortcuts I want, though, and that's when Shortcat is very handy.

Thanks for the suggestion.

As one who's driving Claude daily due to corporate mandates, I can see why people fall in love with genAI coding, but my revulsion has only grown as I've learned to do it, so I won't be spending my free time with LLMs.

Thank you! This looks like exactly what I've been looking for!

The name is very reasonable, but effectively ungoogleable.

I prefer ShortCat's model:

https://shortcat.app/

Similar to Vimium, but for the whole OS. Apparently Homerow is similar, judging from comments I'm seeing here.

I really wish I knew an equivalent for Linux. I might even leave Gnome behind if a different DE has a good model for this.