effect change
8,767 karma · joined July 12, 2010
I sporadically rant at http://masonmark.com and usually see email sent to mason at that domain.
But if you actually have any form of diabetes... definitely do not do that. Unless you are also rocking some other brand. ¯\\_(ಠ_ಠ)_/¯
It's objectively worse for me on every possible axis than Claude Code. I even wondered if maybe I was on some kind of shadow-ban nerf-list for making fun of Sam Altman's WWDC outfit in a tweet 20 years ago. (^_^)
I don't love Claude's over-exuberant personality, and prefer Codex's terse (arguably sullen) responses.
But they both fuck up often (as they all do), and unlike Claude Code (Opus, always), Codex has been net-negative for me. I'm not speed-sensitive, I round-robin among a bunch of sessions, so I use the max thinking option at all times, but Codex 5.1 and 5.2 for me are just worse code, and worse than that, worse at code review to the point that it negated whatever gains I had gotten from it.
While all of them miss a ton of stuff (of course), and LLM code review just really isn't good unless the PR is tiny — Claude just misses stuff (fine; expected), while Codex comes up with plausible edge-case database query concurrency bugs that I have to look at, and squint at, and then think hmm fuck and manually google with kagi.com for 30 minutes (LIKE AN ANIMAL) only to conclude yeah, not true, you're hallucinating bud, to which Codex is just like. "Noted; you are correct. If you want, I can add a comment to that effect, to avoid confusion in future."
So for me, head-to-head, Claude murders Codex — and yet I know that isn't true for everybody, so it's weird.
What I do like Codex for is reviewing Claude's work (and of course I have all of them review my own work, why not?). Even there, though, Codex sometimes flags nonexistent bugs in Claude's code — less annoying, though, since I just let them duke it out, writing tests that prove it one way or the other, and don't have to manually get involved.
If you can't use LLMs to generate placeholder graphics that don't ship in the actual game, then why can you use coding editors that let you use LLMs to generate code?
I'm a programmer, and I enjoyed the sort of "craftsman" aspect of writing code, from the 1990s until... maybe last year. But it's over. Writing code manually is already the exception, not the rule. I am not an artist, and I also really do understand that artists have a more legitimate grievance (about stealing prior art) than we programmers do.
As a practical matter, though, that's irrelevant. I suspect being an "artist" working in games, movies, ads, etc will become much like coding already is: you produce some great work manually, as an example, and then tell the bots "Now do it like this ... all 100 of you."
But banning using AI at all while developing the game is... obviously insane on its face. It's literally equivalent to saying "you may not use Photoshop while developing your game" or "you may not use VS Code or Zed or Cursor or Windsurf or Jetbrains while developing your game" or "you may not have a smartphone while developing your game".
COSMIC isn't yet close to a solution for remote desktop, so I expect to have to wait a few more years. (T_T)
[1]: https://blog.simos.info/how-to-install-and-setup-the-incus-w...
Sorry, but I bought Proxmox 7, but it is not comparable. Incus does everything (and more) with better interface, WAY better reliability, and also not like a hundred EUR or whatever. (100 EUR is fine with me if better, but not if not better...)
I run instances I need to interact with (e.g., do development in containers via SSH and remote-editors, with occasional Remote Desktop) on my very-fast Linux workstation — that also does other stuff like local development, web browsing, etc., but most instances that don't need power run on my old 56-core Xeon enterprise server (used, they are roughly as cheap as a Mac Mini).
Incus makes it super easy to move instances around, and from a skim of the announcement it looks like you could just put Incus OS on some machine you have lying around and drop it into an existing config like that with minimal effort.
I look forward to trying it out, even if my "main" Incus will probably remain on my actual manually-curated Linux desktop.
However, I'm confident that if that is what you want, this is probably fantastic — Incus, including the old LXD (which was mainly built by the same core developers, until Canonical behaved in ways they didn't like, and they hard-forked LXD to create Incus) has been one of my favorite open-source projects for several years.
Fantastic software, steady stream of reliable releases, helpful community... Incus is great.
The main reason work visas are required is not to prevent foreigners from doing work, it's to restrict and control their ability to live in the country, send their kids to school, receive medical care or whatever other social services might be available to residents, or do whatever kind of activity that the government doesn't want people to do and believes foreigners might do (discussing Tiananmen Square, disparaging Charlie Cook, advocating for gay rights, petty crime — whatever it is).
Hiring a foreign company, or foreign contractor, to perform work in their own country is almost always allowed. This is fundamentally not any different than a company having people answer customer phone calls in a call center in Bangalore.
Additionally: it seems likely that it was the result of gas station pump skimmers, just because the card in question had never been used for any other kind of transaction.
If our service goes down for any reason, uh... wait until Monday afternoon, then try again. (Sorry!)
Like, who would die if AOL was down for 36 hours?
Also, when you're 35, you will likely have a much better understanding of what really matters to you, and what kind of partner you really want for the long haul.
It's often said that "60 is the new 50" — and it is, if you do it right — the corollary is something like, "25 is the new 19". That's a gift! Take advantage.
They are doing the work elsewhere, and it would be suicidal for any economy to mandate only using domestic contractors to do all work. (Also, just a side note, even though logically irrelevant: it's not at all the case that these foreigner+robot combos are taking local jobs; I live in Tokyo and have seen that the convenience stores are desperately seeking staff, even 1 day a week, and paying more for it... they just can't find enough (yes, partly because immigration is pretty difficult here).)
For me, though, it's not remotely close. Codex has fucked up 95% of the 50-or-so tasks I asked it to do, while Claude Code fucks up only maybe 60%.
I'm big on asking LLMs to do the first major step of something, and then coming back later, and if it looks like it kinda sucks, just Ctrl-C and git revert that container/folder. And I also explicitly set up "here are the commands you need to run to self-check your work" every time. (Which Codex somewhat weirdly sometimes ignores with the explicit (false) claim that it skipped that step because it wasn't requested... hmm.)
So, those kinds of workflow preferences might be a factor, but I haven't seen Codex ever be good yet, and I regret the time I invested trying it too early.
But not for semantic document structure — recognizing that the grammatically incomplete phrase in a larger font is a heading, recognizing subheadings and bullet lists, tables, etc.
Also not for handwritten text, text inside of images (signage and so forth), or damaged source material (old photocopies and scans created in the old days).
Those areas all seem to me where an LLM-based approach could narrow the gap between machine recognition and humans. You have to sort of reason about it from the context as a human to figure it out, too.
It's kind of qualitatively different from the human perspective, so not a useless concept, but I think that is mainly because we can't help anthropomorphizing these things.
That's a significant difference. There are a lot of tasks that can be done by a n00b with some advice, especially when you can say "copy the pattern when I did this same basic thing here and here".
And there are a lot of things a n00b, or an LLM, can't do.
The study you reference was real, and I am not surprised — because accurately gauging the productivity win, or loss, obtained by using LLMs in real production coding workflows is also not junior stuff.
Although not a pro, or even a good cook, I do like quality knives — but, like most people who develop a casual nice-knife affinity, I then bought a few quality knives, a couple of junk knives to try restoring with a whetstone, a hand-made Japanese knife with my name on it, and... then I had enough kitchen knives to last the rest of my life, and there was no reason to buy any more.
This is the first one I have considered in years. Not because I need it, or because it's necessarily better at cutting must things than the workhorse Sekimagoroku nakiri I use[1] but because it's fucking neat and radiates nerdium waves.
This knife should be celebrated for what it is: basically an epic nerd project. That can also cut at least a few kinds of fruit more easily (or at least more enjoyably).
[1]: KAI model AE5206 (which seems very much like a Shun knife, made by the same company, but cost me only $53)
1.) Enable RDP connections to Wayland sessions, whether they are already running locally, or not (i.e., start a new session if none exist when logging in remotely)
2.) Set that up via SSH, for a remote machine that has no display and anyway is remote so you cannot physically log into it (still very fiddly, but possible)
My requirement is just that every system be remotely accessible via both GUI and CLI. So, RDP (or, theoretically, VNC over SSH would be OK) and SSH.
In the old X11 days, all major Linux distributions met this bar. XRDP worked most everywhere. But Wayland is a very different story.
The only Linux distribution that has Remote Desktop working on Wayland is Ubuttnu 25.04 ("working" per the above, not some "log into the GUI first locally, and then share your desktop" — that almost works in KDE but the experience is very buggy).
The previous editions of Ubuttnu with GNOME almost worked, but logging in remotely would kill any GUI sessions already running locally.
I can still achieve this using X11, so I do. But that doesn't work for my own personal workstation, because I have too many modern (4K or better) monitors for X11 to reliably work. So I need Wayland to drive my actual, physical monitors — and therefore am stuck with GNOME, because I really do need occasional remote access to the entire machine.
I connected to RDP sessions from Linux, macOS, and Windows. (And actually, iPadOS — using Microsoft's app which used to be named "Remote Desktop" but then they bizarrely renamed it "Windows" — leaving me in the rather ludicrous position of saying "I make a remote desktop connection from my Apple iPad to my Arch Linux workstation, using Windows from Microsoft..." ¯\\_(ಠ_ಠ)_//¯
I'm sad because I am stuck with the requirement that all my computers can be accessed via remote desktop (e.g. RDP) in addition to SSH. And I also have to have 3-4 monitors per machine, so I can only use Wayland.
Thus, I am stuck with GNOME on Linux, because no other desktop environment (including KDE) yet has functional remote desktop on Wayland. (Where by functional, I mean equivalent to Windows/macOS where you can log into the same session that may or may not be already running locally.)
I know only 1-2% of users have my problems (^_^) but I just mention them in the hopes that KDE will keep developing krdp and make it work well enough to compete with GNOME and Windows on that axis...
It costs about $100/year per seat for commercial use, IIRC. But it is significantly faster than Docker Desktop at literally everything, has a way better UI, and a bunch of QoL features that are nice. Plus Linux virtualization that is both better and (repeating on this theme) significantly more performant than Parallels or VMWare Fusion or UTM.