pull down to refresh
Isn't meaningful also subjective and qualitative?
I used ChatGPT to help me decipher the first 18 lines of Canterbury Tales:
What that Aprille with his shoures soote,
I appreciate the detailed testimony. You know, I have followed along on your clanker crusades for a little bit now, getting sparse detailes here and there, and I could sense that the vibing scene has amassed huge amounts of bad quality code output that make competent the jobs of competent people (such as you are) harder. There's a shadow of this on the meat space cronyism that I'm more familiar with, which is, as I alluded to above, a false sense of competency. Bad logos, wall-art, signage etc. is everywhere because now everyone thinks real graphic design is obselete. Badly thought out emails with way too much text are sent more often. On net, we get lower quality ....everywhere. And increasingly, fewer and fewer people seem to notice or care.
If compute stopped being so heavily subsidized, and the cost of using it were higher, then my guess is the more meaningful applications would become more apparent.
clanker crusades
I'm mostly crusading against the endless stream of bullshit.
I could sense that the vibing scene has amassed huge amounts of bad quality code output
Yes. And it's tricky to spot for the human eye, but also for vanilla bots. For example, you ask Claude to do a full security review on some vibed crap it will tell you it's great - some minor issues (it will literally flag shit as minor.) But if you actually spec out a quality expectation (that is not even too crazy, like things you would put in a definition of done in a regular commercial software business and that's already super-low-bar) then suddenly there are more issues than the bot is trained to enumerate (really annoying thing they taught these bots to do: limit lists of issues.)
competent people (such as you are)
I guess that depends on who you ask, I'm generally synonymous with rectal pain, lol.
now everyone thinks real graphic design is obsolete
This is exactly the bullshit I'm seeing all around me and it comes directly from the labs pumping their own bags. All the US labs do it and people believe it - even overseas people - because they are dependent on some sociopath giving them hope. It bothers me deeply and endlessly how vulnerable humanity is to bullshit. This is the real issue. LLMs are just tools - poorly developed tools but it's early days - the people blindly repeating false narratives thought up by greedy folks is what makes it dangerous. Very dangerous.
If compute stopped being so heavily subsidized
It'll end only when inflation and salaries catch up to the insane valuations of startups we see, or the hype dies. Not counting SpaceX because that is a conglomerate so it's hard to estimate which of its products will hold, but let's see what happens post-IPO with Anthropic and OpenAI. If these hold the valuations ascribed to them now, subsidized compute will end when overall salaries did a 10x and you pay $50 for a dozen eggs. Then we'll also see if Bitcoin truly is an inflation hedge.
then my guess is the more meaningful applications would become more apparent.
We'll have to see. I do know that there is still good stuff out there, and effort is not gone. It does sadden me to see some formerly okay-ish products become unusable. Especially in nostr this is now prominent, which sucks.
Improvement is a relative and subjective thing, so it will be different for everyone. I think I'm a little better than break-even overall, but it's negligible in the big picture right now. New problems have come that require new solutions, I'll give the worst example from real life.
For years I have been reviewing the code I run on my "batphone". I was losing this battle, especially by mid 2024. For example, I didn't have email on that phone (I still don't, but by choice) because every time I was halfway reviewing the Proton mail app they released new code - I was continuously behind. I could barely review Signal within their cycle. The asymmetry of a bunch of teams pushing out code versus a solo reviewer was slowly eating up all my time, also because making sense of code snippets consumes a lot of energy. I ended up spending 4 days a week on security review alone, just to review OS changes and 5 apps. I was losing track of what I had and hadn't reviewed, making a mess of my reviews. It was a never ending downward spiral and spending time on process improvements meant potentially falling behind on important security updates.
It got worse through 2025 as some apps and more importantly, the underlying libraries they are a composite of, started to be vibe coded. The review pressure did a 4x or so (and I started seeing more and more problematic things.) So something had to change: either I give up knowing what code I run or I find a more efficient way to do the reviews. I could hardly reduce apps further and I was getting more and more worried about my day-to-day phone based on what I was seeing. I needed to increase scope, not decrease.
This is when I decided I must try to "fight fire with fire". I just had to make sure I have superior fire: faster and more precise than the fire I'm fighting. Luckily for me, most people that call themselves devs don't care one bit about quality. And precision is intensive when using LLMs, especially if you're too lazy to read the results - you cannot vibe review, it's retarded to do so because the (post-)training data is too poisoned by lazy and retarded people operating on "good enough" for it to be reliable.
So, I went about describing how I do security review for software and build that out to make specific instructions for each app, a generic framework that detects new releases and new or changed dependencies, and tooling, tons and tons of tooling, because everything that can be done deterministically, should not be done by an LLM. I have this now and I have been able to expand the scope to my non-batphone. I also no longer spend more than 8 hours a week on review, except when there are some completely insane changes.
The process I sovereignly have is about the same as what socket security does, except I have 10x the depth, 20x the precision, -10x the false positives (and I know about them early) and nothing a bot says will ever be published to anyone except to me so I don't add to the global accumulation of slop on the internet. I don't get a final report, only intermediate results. Sometimes I find something worth proposing a fix for, but devs have gotten so insensitive to input due to all the bots that both roam GitHub and that they choose to interact with, that it takes forever to get even low hanging fruit acknowledged, let alone merged. So I am no longer spending the effort to send people high quality pull requests. I don't send them any pull requests; I just either remove the code, or I fix my fork and give the world the finger, because no one cares about quality anymore.
As of mid 2026, nearly everyone, including the main LLM labs, is hyper-focused on the act of shipping and is using LLMs to do it. It's hard to find a dev that truly cares about quality nowadays: everything is "good enough" by a very, very low bar. Coding LLMs made this possible to even the biggest retards now through vibe coding but I've seen formerly smart people become retarded now that they can be lazy. It's the direct enshittification of software and most of all FOSS, because that doesn't make many people money.
For commercial products, maybe a better word would be
Theranosification. Faking till making. It's all around us in densities never seen before. And most of the time, software devs don't even understand that the shit they pitch is completely fucked, because they didn't read the slop they published, let alone understand it, and then you get some asshole pitching their slop as if it is the newest bestest thing to ever been made. They couldn't be further off the truth.Can LLMs make things better? Yes.
Will you work less? No.
You will work harder and solve more difficult challenges.