hombre_fatal

Lots of small fs operations on macOS is apparently rough.

I use nix on macOS and any time it's making changes to /nix/store, the Finder and fsevents processes peg cpu.

IIRC his video shows fs-heavy ops like file deltions and pnpm install taking much longer on a macbook despite better specs.

Frankly, things like the lack of a universal cmd-c/v clipboard on linux has been 1000x worse for usability than slower bulk fs.

There's some hyperbole but his point is obvious to me from hanging out in forums and IRC channels over the years, or reading snarky "umm, I'm confused...?" quote-tweets for certain Omarchy posts about how Omarchy is just quickshell hyprland rice.

I used to use an old omakase version of Debian called Crunchbang in uni (https://crunchbang.org/) and there was always someone around to make the remark that you'd only use it if you couldn't read manpages.

Or Dropbox seems pointless because you can build it trivially by mounting an ftp location with curlftpfs + svn.

For a single document like a blog post it should be easy, though you don't have to sub a famous author in for X.

There are all sorts of ways to do it though, like pointing it to an example doc.

He's talking about Linux neckbeards and gatekeepers, not, say, cultural idiosyncrasies of any given distro.

Wow, what a great demo video. And this is the worst this tech will ever be. Though maybe nobody will ever beat the dog's hilarious persona.

I don't see why Anthropic would be expected to publish default Claude voice or use Claude for their public writing. No matter how great AI is, it doesn't commit you to using it for everything.

And they probably know basic LLM tricks like "write it in the style of X".

When you read a blog post with Claude voice, you're seeing the result of someone who couldn't even be bothered to do that which is why they deserve extra lashings.

But slowing down to decipher text instead of single pass simple reading is the problem.

Random example from a session I have open:

> *One new failure mode the doubling opens, and how it is closed.* A ring never shrinks. With two 8-byte index words per record instead of one, the ring's doubling comes inside the budget's reach in the blank regime: a doubling taken [...]

It takes extra deciphering cycles to see "the doubling", "budget's reach", "the blank regime", etc. and figure out what it's referring to. I had to read that opening sentence multiple times. At first I parsed it like "One cat the table yawns".

Sometimes it's so encumbered I can't tell what it's saying at the directional level: good or bad, fast or slow? "Your blank regime negated the pre-armed run's dilemma but clawed back the overall metrics."

It's not how I'd phrase things if I were trying to be easily understood, though Claudese is probably great for LLMs due to ad hoc jargon usage.

> because that's how it advertised.

But you didn't provide the evidence for that. You shared some links and then admitted they didn't claim it.

It kinda seems like "because I think they're a little too positive about their product, I can set my expectations to anything I want and la-la-la it's their fault."

And I don't see the problem with agents building test scaffolding as they go. It might be too defensive at times, like testing a shell script you don't run often, but big deal. It's kinda cool imo, and it's trivial to make it stop.

Seems like motte and bailey fallacy. They say their models are good (the motte), therefore their models must one-shot everything to perfection (the bailey).

Besides, other people's claims about something doesn't give you license to abandon all critical thinking. Though it's evident they don't claim what you say they are.

I wouldn't agree. Sota models can do self-directed sampling, profiling, benchmarking, read call trees, etc. to give you a report of the app's bottlenecks and then recommend solutions that can be vetted.

I do this constantly.

As the upstream comment points you, you don't need to specify. Sota models are that good. And by being overprescriptive you can accidentally shut off branches that they would've taken, downgrading the quality of their work.

These remarks about how they'd agree with you if only they'd read more or how they aren't as superior to others as they think, these make your comment worse and emotional. Just make your case like you would in any other comment. If it's persuasive, then you look right and they look misinformed.

This kind of response is why we can't talk about all sorts of hot button topics on HN: people get personal. I'm sometimes guilty of it myself.

Impressive, wrong on both counts.

(1) HDR display (e.g. you bought a macbook/iphone) doesn’t mean you want random videos in the instagram feed to be full brightness despite setting your screen brightness to minimum in a dark room and (2) reducing white point just darkens stuff across the display, it doesn’t disable the thing people are complaining about.

You put more heart and emotion into your comments here than I spent picking a non-Grok AI service.

I think what makes that so annoying is having to backtrack which is super jarring.

One idea is letting you press enter to jump to the start of the next word and immediately resume.

LLMs trivially catch these things in review, and if you find agents agreeing incorrectly with each other (weak models?), maybe you need to be doing adversarial review.

It's pretty much solved, though I only use sota models.

I think for these convos, we need to see concrete fail cases so we can see what you're talking about and whether the truth matches up with the claim.

You say it's "letting politics eat your life" yet we're talking about alternative services that are all very similar. Clicking a different website's checkout button is to consciously decide who benefits from your decisions. I think that's how you should live -- pick the oat milk too.

There are other reasons to not use Elon Musk's AI beyond politics, though this month he's investing $200 million to sway the Texas midterms back to an awful politician, far closer to home than what Xi Jinping might be doing.

It wasn't long ago that Elon tried to bias Grok to not say bad things about him to the point that Grok would say that Elon was the best piss-drinker in the world and that Elon was in the top three of every category (basketball, mathematics, physics, etc). https://newrepublic.com/post/203519/elon-musk-ai-chatbot-gro...

If I have to tie-break between competing AI services, I'm hella not choosing that one.

That said, performance and security considerations can be automated by a reusable harness and I think within six months it will just be built in to things like claude code and codex the same way you might tack on a performance/security section to your project's AGENTS.md.

A crude example is how, when building a large forum project years ago, I had a checklist I ran on every new http route: authentication, authorization, validation, etc. just because it's trivial to forget one of them. Yet now the LLM can come up with a lintable/verifiable way to ensure every route has the whole kit by construction.

Almost every way I feel like I can provide unique value to the automated AI workflow (aside from deciding what to build at all, and taste) feels like it can be automated by better markdown.

This starts to feel like you're defining the word intelligence out of any meaning and out of any way we apply that word.

So when LLMs can do all human knowledge work, and do it better than humans, we'll be in the mines listening to you go on about how it's actually just autocomplete or just math, a distinction that apparently means nothing.

Well, it depends on what you want. I'm glad the Claude Code TUI runs in my terminal, for example, because that's where I do software development and interact with the file system and run other lightweight keyboard-driven TUIs that I want to use moment to moment.

Imagine if there were no terminal/pty system. Then what, I'd have to use Claude Desktop (doesn't support multiple windows) and Finder?

So no, it's not the only argument for TUIs.

Good point. "decisions.md" is a good starting spot.

The point, of course, is that you need some way to document decisions and pivots and eurekas (including rejected ones) over the lifetime of the project. It's a key part of invariant discovery and then committing to them.

The context here is using AI agents to automate high-quality software development, something that is slow and expensive to do by hand which committed us to suboptimal maximas like "that subsystem is too gnarly and fragile to improve, so we won't" and "we used React Native in the Windows start menu because it's what the developer knew best".

I don't understand what you're responding to.

TUI apps are multi-instance by default since the user can invoke the program in different pty sessions. The developer would have to actively fight that.

GUI SDKs generally require the developer to pick a solution. And it's easier to default to the simpler solution where you have one instance, one data model, one window.

Whether things might change in the future isn't relevant because we're talking about how things are today, not speculating about the future. And the time to change one's opinion is when reality changes, not ahead of time based on how you think things hopefully will be.

Then you'd have to contend with the rest of the list of TUI advantages over GUIs, like trivially operating them as they run on remote machines.

GUIs have their benefits. You can make much richer UIs with much nicer UX when you're not limited to a grid and whatever paradigm you can hack into the ancient tty stuff, and when you have access to raw keyboard scans.

But I don't think I've run into a TUI that I couldn't have multiple instances of.

Anything from WhatsApp to the Settings app on macOS where I might want to, say, compare the wifi network settings for two access points.

But the point is that the developer has to build the app in a way that lets you do this.

Once again, there is no default with the tools I listed. You decide on how you want windowing to work and then build it.

At this point, show me examples of needing a mutex to "block" a multi-window default.

They don't do anything by default. The developer makes it do what the developer wants (not the user). You claimed the developer has to "actively block it" which isn't the case.

Even in SwiftUI which has the simplest way to support multiple windows with zero work, the developer has to use WindowGroup, and if they didn't then you don't get to open more than one window.

Which desktop?

It's not true for AppKit, SwiftUI, GTK, Qt, WinUI, or Electron, and I'm not sure what's left.

One of the biggest upsides of TUIs over GUIs is that I can run any number of TUI instances.

Meanwhile a GUI's developer has to decide to grace me with the ability to even open more than one window. "A tabbed interface will be sufficient!" -- yay, I'll never be able to view two screens of info at the same time.

I use Claude model=fable effort=high to write most of my initial plans, especially the hard things like re-architecture.

I recommend putting time into figuring out the "soft infrastructure" of your projects, usually things you've never been arsed to do in your life (like me until this year).

Start with a docs/*.md folder that you link to from AGENTS.md where you can document things that would be important for you, any human, and especially AI to know. Start with a super basic ADR system where you document invariants over time, maybe a single DESIGN.md file that just lists them. Have a references/ directory that shallow clone all of the important deps you use for local agent reference (and tell the agents about them in AGENTS.md). Keep your plan files to higher level things like invariants, decisions, accepted risks, rejected ideas, that way you review the hard stuff, and sota models these days write crazy-good code.

Look at using AI as a skill that you have to learn and develop, not like how we see CSS where it's the technology's fault if we suck at it for life while putting zero effort into developing the skillset.

Or maybe there's some load bearing "make impossible states unrepresentable" platitude in my AGENTS.md I'm overlooking that's responsible for why I have such a great experience with AI and nobody else on HN apparently is.

I've built many things this year letting sota models design the architecture, and they've been well-reasoned about it.

I think what you describe is a lack of harness/infra/process around working with agents on a project. No effort to document ADRs. No effort to document invariants. No effort to design processes that ensure maintainable software even if it were five years ago and only humans were writing the code.

Just because you delegate important decisions to AI doesn't obsolete practices as trivial as maintaining a log of potential issues in your codebase that you'd want to work through, keep track of, and verify.