saltcured

It smells like delusions of grandeur, to be so shameless about asserting his own enterprise as the highest priority for society.

Whether he truly has these beliefs or just cynically thinks it is a good way to message, it disturbs me. It only makes it worse, seeing how quickly these new platforms can invade and contaminate the fourth estate (or the mass-market communications that have essentially subsumed that role already).

I think the person above is stating that this is distinguishing characteristic of "good" developers. By extension, the ones you are referring to are not good.

For me, the "P" key is right at the split between zones. I probably (nearly?) always type "O" with my middle finger. But, the "P" is easily done with middle and ring fingers, and it seems like I may split the duty in many words.

It's hard to be sure, because I'm not sure how much I bias my movements by trying to monitor them while I do it! Normally I just type on auto-pilot, looking at the screen and not really being aware of the specific movements or finger duties...

I have a very idiosyncratic touch-typing method. I am self-taught with 40+ years of computer usage since grade school. There are arcing zones that are in the natural reach of different fingers, but with overlap where adjacent fingers can reach the same keys. I think my middle fingers are somewhat dominant, and may do more than their share of the total keypresses because of their dexterity and reach.

I have relatively large and flexible hands, so I span the keyboard by bending and reaching fingers rather than moving my forearms. The outer heel of my hands are statically parked on the palm rest or desk, and just the small joints in my hands and fingers are moving.

This parked posture is probably why I strongly prefer the Thinkpad "trackpoint" input device for general computer use.

Yeah, thanks for that. I can only imagine someone reacted as if I am arguing against solar, which I'm not. It's just that you really need storage to phase out the natural gas peaker plants, since they need to deliver on-demand rather than as-available.

And, this gets even worse if you are also replacing baseload plants with solar, since the solar output is more variable than those old plants were, so now you need even more storage capacity to offset those dropouts at the same time you are handling demand peaks.

Sometimes, I think there is some magical thinking going on when people see solar panels work at a house without batteries, and they fail to recognize how much they depend on the grid for stability. If they ran in island mode, they'd quickly understand just how important storage is...

And, for larger undergraduate courses, the graders may be undergrads hired to do this menial work for the TAs, while the TAs focus on things like extra study sections and office hours.

As a CS undergrad, I did that kind of work at UC Berkeley in the 90s, grading courses that I'd taken 2-4 semesters earlier. I even did a semester as a TA, because the professor was having a hard time finding enough grad students who wanted to deal with one of those larger classes.

There seems to be an analogy here between these kinds of programmers and the many types of "machinist" that were coded in earlier economic studies.

But, these are job categories, not separate industries. I think the industry category would be more about the employer or the product category, not the individual job codes of their employees..?

Whatever it is doing, it won't be doing it for very long ;-)

Hmm, how much of that is base load versus the dramatic rise of natural gas plants that respond to dynamic demands? Even in daytime, I think you need storage to get rid of those, right?

Hah, from the headline here, I was expecting a completely different article about traditional file management UX, i.e. before the cloud UX of "it's always saved, except when it isn't" started to invade everything.

Not only the deliberate action, but the period with progress and success indicators, since this operation took a while and could go wrong along the way.

Yeah, I added a boring case on day 1 and would have trouble describing the feel of the bare device right now.

Found one on Amazon so I dodge Google's exorbitant plastic markup, and it is a relatively minimalist wrap that smooths out the camera bump and makes it more grippy in hand, in pockets, and on flat surfaces.

The behavior of those slippery phones in pockets is almost the bigger gripe to me. Like they are a new class of motion amplifier, trying to rotate and squirt their way out at the slightest twitch.

I think your thought experiment is begging the question by asserting the outputs are indistinguishable.

The poster above is complaining that the actual outputs are qualitatively "not creative". That they embody a statistical norm, while creative work has more unique variance.

These two viewpoints cannot be bridged by a flourish that ignores the fundamental difference of opinion...

Yes, that's why I left wiggle room for a few choice packages you track yourself.

What isn't really tenable for a small team is to drag in scores to hundreds of different third-party packages and rely on them all when you have no staff allocated to really evaluate and monitor so much upstream stuff.

The mitigation is to ignore nearly all that stuff in PyPI, realizing it is barely a step up from copy-pasting out of stack overflow. Same goes for, e.g. python-whatever RPMs that look like they were drive-by contributions.

So your trusted base is the Python runtime and maybe a few choice libraries that actually have engaged package management. Sadly, it is a trap to think you can put things on the internet as a small player by leveraging lots of third-party stuff, when you don't have the resources to actually maintain that large dependency base.

This also may mean you want a base OS like Fedora and not some crusty "enterprise" variant that tries to bundle a frozen Python from 15 years ago. This is getting a little better in recent years, with some enterprise platforms having optional package streams for parallel install of newer runtime versions, etc.

Edit to add: on the Fedora-derived platforms, pip doesn't really interfere much anymore. The platform python-* RPMs go into /usr/lib/... while pip installs into /usr/local/lib/... or your ~/.local/lib/... hierarchy. The only mysterious behavior you get is if you manage to install in both places and then you have to understand the path search order. Just like if you do strange things with user and virtual environments.

Heh. This is part of why I always preferred getting my python runtime and all these core libs from my linux distro packager. But python die-hards look at me like I'm insane when I suggest this.

Even in today's LLM-addled world, I still think it's a better tradeoff for smaller system/site operators. Get the distro-level testing and regular auto-patching, rather than telling yourself it is a good idea to "freeze" your python world at the pace of your slowest app/system integrator.

It's the same reason I think docker is an anti-pattern for these small groups. You really cannot safely use these strategies to bake an image and divorce yourself from the maintained distro layer unless you are going to adopt CI/CD that does the same thing. Keep tracking all these patches to the entire runtime with all the dependencies, rebuild, re-test, and redeploy.

It's total cargo culting to adopt the freeze/image techniques and just abandon apps in place as if that is robust!

But, it's also more than just automatically updating everything. There is an editorial aspect to distro management to help secure the supply chain to your app. Most smaller orgs cannot feasibly do this as well as the regular linux distro maintainers. Some large orgs can probably do it better, but that doesn't matter if you aren't doing it large org style with dedicated security folks tracking all this against well-developed policies and procedures...

To be fair, Java applets and the Flash plugin socialized that before Javascript tried to walk us back from the edge just a bit.

And before that, whatever Microsoft "macro" or embedded code variant was popular that week in all those document formats...

Visual imagination and inner monologue are not complements, but more like orthogonal capabilities. I am pretty far away from both. Sorry, for the long reply, but this topic hooks me every time I trip over it... ;-) Ironically, given what I will state next, people around me tend to describe me as very imaginative.

I lack conscious visual imagery. I can't recall faces of loved ones or really "see" scenes from memory in a way that feels visually coherent. Even though I have practiced drawing and painting at times, I cannot render people, animals, nor landscapes from memory. I would need a live model or reference photo. I could draw or paint fictional scenes or more basic forms, but would struggle with anything life-like without a reference. I can recognize it is flawed, but cannot summon the missing bits.

I also lack any inner monologue. My thinking is not verbal. It is extra work I have to do to cast ideas into words after I've already thought them up. I can think about words as a task, but I don't think in words. This interviewing strategy also sounds like a nightmare to me. I could talk at length about "how I think" in some philosophical or psychological mode, but I cannot recite a real thought process as if it is a sequence of verbalizable intermediate steps. That's just absurdly ill-fitting to me.

By contrast, I have a pretty strong inner stage for proprioception and a sense of spatial relationships or topology for both real 3D and 4D (space-time) structures. If I focus, I can nearly invoke physical sensation by imagining falls, spins, impact, flight, etc. This can also involuntarily engage my inner audio stage, where I might imagine corresponding sound effects.

There is a first-person and third-person version of my spatial stage. I can sometimes think about a dynamic stage, i.e. the 3D stuff moving, but more often this turns into a more static sense of a stage that includes motion vector information. (Alas, this hasn't made me a dancer nor given me appreciation for choreography...) I don't know how to describe this to people though. I don't "see" it. I feel or know it. It may have a perspective-like projection to it, favoring or enlarging the near space, but it doesn't have occlusion, lighting, or other visual qualia.

Sometimes, but not always, I cast more abstract thinking onto topological structures that map into this spatial stage. Then I can "feel" the shape of a problem, problem space, or search process over that.

As I mentioned above, I do have an inner audio stage. I can think about arbitrary sound and it has an experience like a faint version of real-time hearing on a secondary channel. There are intensity and frequency qualia, timing and rhythm, etc. It can convey harmony, noise, sharp attacks, etc. At the extremes, I can nearly induce accompanying physical senses like a deep bass thumping my chest or a high-hat sizzling in my ear. But it's always sort of off in the corner as an apparatus I can access, without being mistaken for real hearing. This and real hearing can operate in parallel, though it is harder to attend to the inner stage if real hearing is overwhelmingly loud or full of semantic content.

When I think about words, it is sometimes in this audio stage. To some degree, I can remember people talking or run through hypothetical speech with intonation etc. It's a bit thinner on the audio qualia than other audio thinking, as I focus more on the semantic and emotional content and less on the sounds qualities. I know whose voice I am imagining, but it's not necessarily a rich or faithful synthesis of the sound qualia. Musical thoughts bring more of that richness.

But, I have another linguistic stage that is more symbolic. It's more like feeling the topology of the words, a bit like a parse tree, and less a real-time stream. It would be a mistake to say it is a visual process, as it morphs into semantic units in my spatial/topological stage. It can also abstract, thinking about the sequenced structure of sections, chapters, etc. rather than all the words.

Finally, all of the above-mentioned things feel like tools I can apply. They don't seem like "me" or my core thoughts. It is very difficult to describe my thoughts that happen behind the scenes in all this. There is a version of the above senses that represents my real, anchored, position in the world. And there are the extra, "imagining" versions that I can operate. My real self is some amorphous attention that flits among these.

(Probably) due to trauma effects, I have very high situational awareness. So, it's very rare that my attention can completely separate from the real-world monitoring. But I can deprioritize it a bit when I really focus on other problem solving, recall, or rumination. It's a bit like a mixing board that cuts from one blend to another with different levels of abruptness versus smooth blending or transition...

Thanks. I worried that Fleisch one might be just what you said..

Does Wurstsatzwerfer sound to you like a pun on Ersatz + Werfer? I guess that might not work, and also isn't as funny in German. In English, the borrowed word "ersatz" brings a connotation of cheapness or fakeness.

Hmm, would any of these lead to more immediate understanding and laughter?

Der Wurstsetzwerfer

Das Fleischgeschützer

Der Wurstbeschleuniger

I think the traditional scientific use is still future-oriented. The hypothesis-as-prediction is a forecast about the result of an implied, future test.

I'd argue that stats/probability folks pervert "prediction" by using it in contexts of mere information masking, stripping away the tense. There are many better words for this: speculate, extrapolate, estimate, infer, ...

I think in normal English of the public at large and lay people, "predict" and "forecast" have nearly the same meaning as speculation about a future outcome. The main difference is that we "predict" discrete outcomes while we "forecast" statistical outcomes.

This argument is conflating determinism as in "byte-for-byte reproducible builds" with determinism as in "builds are semantically correct".

The fact that a traditional PL compiler output can have different performance, based on stochastic optimizations, is not in dispute. But, we consider the compiler flawed and inappropriate for use when it will randomly produce incorrect machine code. Under normal circumstances, we can infer that "correct" source code will produce "correct" builds every time it is fed through a valid compiler.

The problem people have with slop and vibe coding is that there is no formal semantics to even define correctness of the LLM/agent output.

With no formal semantics to constrain the outputs, you have to treat each prompt/inference run as a fuzzy search. What it "finds" may or may not match your intended semantics. It is merely a candidate that might correspond to your ambiguously specified semantic intent.

It is magical and lazy thinking that leads people to use this fuzzy search process in "I'm feeling lucky" mode.

Presumably, they still want the free tier to help acquire new customers?

You could say they should separate the human users from the slopsters in any tier. This could be done with some quota/rate limiting model.

But, if they are either on the fence or believing that agent-based development is the way forward, then that new reality is the one they would need to support. Cutting it off will send those customers elsewhere.

And, if you've managed to handle the new workload, it is basically in the noise to also handle artisanal human developers in the background on the same solution...

Left this way, every collaboration can turn into what used to be the worst nightmare type with that one unavoidable stakeholder or coauthor who likes to procrastinate and then create an insane fire drill in the final moments before a deadline.

Now, no matter the amount of preparation work, those last rounds can completely rewrite something with no hope of review. No iterative improvement. No ratcheting towards a known quality. Just a bunch of cargo cult review followed by YOLO-style, vibe-everything absurdity.

I used to think it was a struggle to walk the user through introductory EM physics.

But, it turns out that was a walk in the park compared to explaining how to acquire and isolate the dopants, not to mention building up the pure silicon wafers.

I'm not sure it maps well to game theory. I think it's more of a false dilemma.

There is the old adage in many academic fields that "progress happens one funeral at a time." It's an age-old, rueful take on how our own learning and adaptation proceeds within a collective society. It is a rendering of an objective truth seen at the intersection of evolutionary processes, memetics, and the human condition. It is not a call to action.

Sometimes, I think the most worrisome thing is the ever so slight possibility that some clever, naive upstart will hear this adage and, rather than ponder the lesson about patience and complex social systems, instead conclude that the short path to a lot of progress is a bunch of killing.

Even if we ignore the obvious psychopathy it takes to embrace such an idea, it is still fantastically myopic. Yes, in a hypothetical way it might happen... after the great chaos, some new civilization might emerge with an impressive growth spurt.

But, it is infantile fantasy to believe that such chaos can be applied in a goal-directed fashion, or that the perpetrator(s) can expect to survive it and enjoy the result. Nobody can control who nor what knowledge passes through the gauntlet, nor what direction it spins after that.

What troubles me most, perhaps, is that the current zeitgeist makes me think some people are oblivious to these caveats.

> Honest question: What is preventing you from understanding any part of the stack now?

Bringing this question out of the thread and back to the current moment...

the recurring answer now is "futility in the face of churn", I think.

I completely disagree with the way you frame this.

The "abundance" is not all positive valence. It is a flood of superficial good with an unmanaged ratio of negative byproduct mixed in.

Some of the "change" is not merely doing things a bit differently. The LLM flood is effectively attacking existing feedback and control mechanisms without any attempt to establish appropriate replacements.

I think there are several more dangerous sentiments coursing through civilization these days:

1. That people bear no moral responsibility to act in good faith, and the corollary that one can enjoy the fruits of production while externalizing all the liabilities.

2. That strategies like "move fast and break things" and "fake it until you make it" are universally applicable and appropriate, rather than only in the low-stakes entertainment/media/marketing fields where they flourished in recent decades.

3. That mere acquisition of capital gives one the right to perform N=1 experiments on global civilization, the economy, and the enviromment.

I think the tricky part is if they start cranking up the intensity to counter it being scanned sequentially over a larger area...

First, what is the fail-safe in case the scanning mechanism fails? I.e. is the intensity safe if the beam becomes stationary? Or does it guarantee it goes dark when not properly scanning?

Second, what is the safe intensity limit for the brief visits to each target pixel area? As the scanned area increases, you need higher intensity to maintain the same visual brightness with a shorter exposure for any one spot. At some limit, is the pixel to area ratio too low and this intensity too great?

Third, does the optical path through the cornea and lens have any hot-spot where you have to also worry about peak flux with this projection technique?

I think you could use AI for this if you do it properly as a sort of adversarial coding project. Have it build tests to break the code. Don't give it the job of making a test suite that passes.

I think people higher up are warning against a naive mistake, which is asking one agent to write both the product and the test suite. Here, making things up and cheating becomes a problem of quality theater...

I tried to spelunk this thread and couldn't find the topic I want to see explored.

I don't really have the crypto chops to declare a fact here, but I have a speculation or intuition. In this day of supply chain worries, I think a proper signing algorithm should not be signing this tower of hashes, or not just this tower.

It should incorporate a canonical stream of all the actual commit content. It is the integrity of this content from the author's working copy that they can and should attest, not some derived byproduct of the storage scheme. Edit: Of course, I mean a secure hash of this stream, not a signature including a copy of the entire content!

My intuition is that the content-addressable store is used to reconstitute the commit content, but the verification should be over the original content, not the internal addressing of the store.

Wouldn't this make it harder to do these exploits? You would have to find alternate content that simultaneously produces collisions in the internal addressing hashes and for the overall canonical stream hash.

If you also carry size info alongside each hash, would this also make it much more difficult to produce useful collisions?

The statement further up should have qualified it as "scarce commodity". Then it is correct.

To successfully perform dumping, you have to have access to excess supply capable of disrupting that market.