My father, also known as “R0ML” once described a methodology for evaluating volume purchases that I think needs to be more popular.

If you are a hardcore fan, you might know that he has already described this concept publicly in a talk at OSCON in 2005, among other places, but it has never found its way to the public Internet, so I’m giving it a home here, and in the process, appropriating some of his words.¹

Let’s say you’re running a circus. The circus has many clowns. Ten thousand clowns, to be precise. They require bright red clown noses. Therefore, you must acquire a significant volume of clown noses. An enterprise licensing agreement for clown noses, if you will.

If the nose plays, it can really make the act. In order to make sure you’re getting quality noses, you go with a quality vendor. You select a vendor who can supply noses for $100 each, at retail.

Do you want to buy retail? Ten thousand clowns, ten thousand noses, one hundred dollars: that’s a million bucks worth of noses, so it’s worth your while to get a good deal.

As a conscientious executive, you go to the golf course with your favorite clown accessories vendor and negotiate yourself a 50% discount, with a commitment to buy all ten thousand noses.

Is this a good deal? Should you take it?

To determine this, we will use an analytical tool called R0ML’s Ratio (RR).

The ratio has 2 terms:

1. the Full Undiscounted Retail List Price of Units Used (FURLPoUU), which can of course be computed by the individual retail list price of a single unit (in our case, $100) multiplied by the number of units used
2. the Total Price of the Entire Enterprise Volume Licensing Agreement (TPotEEVLA), which in our case is $500,000.

It is expressed as:

1

RR = TPotEEVLA / FURLPoUU

Crucially, you must be able to compute the number of units used in order to complete this ratio. If, as expected, every single clown wears their nose at least once during the period of the license agreement, then our Units Used is 10,000, our FURLPoUU is $1,000,000 and our TPotEEVLA is $500,000, which makes our RR 0.5.

Congratulations. If R0ML’s Ratio is less than 1, it’s a good deal. Proceed.

But… maybe the nose doesn’t play. Not every clown’s costume is an exact clone of the traditional, stereotypical image of a clown. Many are avant-garde. Perhaps this plentiful proboscis pledge was premature. Here, I must quote the originator of this theoretical framework directly:

What if the wheeze doesn’t please?

What if the schnozz gives some pause?

In other words: what if some clowns don’t wear their noses?

If we were to do this deal, and then ask around afterwards to find out that only 200 of our 10,000 clowns were to use their noses, then FURLPoUU comes out to 200 * $100, for a total of $20,000. In that scenario, RR is 25, which you may observe is substantially greater than 1.

If you do a deal where R0ML’s ratio is greater than 1, then you are the bozo.

I apologize if I have belabored this point. As R0ML expressed in the email we exchanged about this many years ago,

I do not mind if you blog about it — and I don't mind getting the credit — although one would think it would be obvious.

And yeah, one would think this would be obvious? But I have belabored it because many discounted enterprise volume purchasing agreements still fail the R0ML’s Ratio Bozo Test.²

In the case of clown noses, if you pay the discounted price, at least you get to keep the nose; maybe lightly-used clown noses have some resale value. But in software licensing or SaaS deals, once you’ve purchased the “discounted” software or service, once you have provisioned the “seats”, the money is gone, and if your employees don’t use it, then no value for your organization will ever result.

Measuring number of units used is very important. Without this number, you have no idea if you are a bozo or not.

It is often better to give your individual employees a corporate card and allow them to make arbitrary individual purchases of software licenses and SaaS tools, with minimal expense-reporting overhead; this will always keep R0ML’s Ratio at 1.0, and thus, you will never be a bozo.

It is always better to do that the first time you are purchasing a new software tool, because the first time making such a purchase you (almost by definition) have no information about “units used” yet. You have no idea — you cannot have any idea — if you are a bozo or not.

If you don’t know who the bozo is, it’s probably you.

Acknowledgments

Thank you for reading, and especially thank you to my patrons who are supporting my writing on this blog. Of course, extra thanks to dad for, like, having this idea and doing most of the work here beyond my transcription. If you like my dad’s ideas and you’d like to post more of them, or you’d like to support my various open-source endeavors, you can support my work as a sponsor!

What’s a good maximum line length for your coding standard?

This is, of course, a trick question. By posing it as a question, I have created the misleading impression that it is a question, but Black has selected the correct number for you; it’s 88 which is obviously very lucky.

Thanks for reading my blog.

OK, OK. Clearly, there’s more to it than that. This is an age-old debate on the level of “tabs versus spaces”. So contentious, in fact, that even the famously opinionated Black does in fact let you change it.

Ancient History

One argument that certain silly people¹ like to make is “why are we wrapping at 80 characters like we are using 80 character teletypes, it’s the 2020s! I have an ultrawide monitor!”. The implication here is that the width of 80-character terminals is an antiquated relic, based entirely around the hardware limitations of a bygone era, and modern displays can put tons of stuff on one line, so why not use that capability?

This feels intuitively true, given the huge disparity between ancient times and now: on my own display, I can comfortably fit about 350 characters on a line. What a shame, to have so much room for so many characters in each line, and to waste it all on blank space!

But... is that true?

I stretched out my editor window all the way to measure that ‘350’ number, but I did not continue editing at that window width. In order to have a more comfortable editing experience, I switched back into writeroom mode, a mode which emulates a considerably more writerly application, which limits each line length to 92 characters, regardless of frame width.

You’ve probably noticed this too. Almost all sites that display prose of any kind limit their width, even on very wide screens.

As silly as that tiny little ribbon of text running down the middle of your monitor might look with a full-screened stereotypical news site or blog, if you full-screen a site that doesn’t set that width-limit, although it makes sense that you can now use all that space up, it will look extremely, almost unreadably bad.

Blogging software does not set a column width limit on your text because of some 80-character-wide accident of history in the form of a hardware terminal.

Similarly, if you really try to use that screen real estate to its fullest for coding, and start editing 200-300 character lines, you’ll quickly notice it starts to feel just a bit weird and confusing. It gets surprisingly easy to lose your place. Rhetorically the “80 characters is just because of dinosaur technology! Use all those ultrawide pixels!” talking point is quite popular, but practically people usually just want a few more characters worth of breathing room, maxing out at 100 characters, far narrower than even the most svelte widescreen.

So maybe those 80 character terminals are holding us back a little bit, but... wait a second. Why were the terminals 80 characters wide in the first place?

Ancienter History

As this lovely Software Engineering Stack Exchange post summarizes, terminals were probably 80 characters because teletypes were 80 characters, and teletypes were probably 80 characters because punch cards were 80 characters, and punch cards were probably 80 characters because that’s just about how many typewritten characters fit onto one line of a US-Letter piece of paper.

Even before typewriters, consider the average newspaper: why do we call a regularly-occurring featured article in a newspaper a “column”? Because broadsheet papers were too wide to have only a single column; they would always be broken into multiple! Far more aggressive than 80 characters, columns in newspapers typically have 30 characters per line.

The first newspaper printing machines were custom designed and could have used whatever width they wanted, so why standardize on something so narrow?³

Science!

There has been a surprising amount of scientific research around this issue, but in brief, there’s a reason here rooted in human physiology: when you read a block of text, you are not consciously moving your eyes from word to word like you’re dragging a mouse cursor, repositioning continuously. Human eyes reading text move in quick bursts of rotation called “saccades”. In order to quickly and accurately move from one line of text to another, the start of the next line needs to be clearly visible in the reader’s peripheral vision in order for them to accurately target it. This limits the angle of rotation that the reader can perform in a single saccade, and, thus, the length of a line that they can comfortably read without hunting around for the start of the next line every time they get to the end.

So, 80 (or 88) characters isn’t too unreasonable for a limit. It’s longer than 30 characters, that’s for sure!

But, surely that’s not all, or this wouldn’t be so contentious in the first place?

Caveats

The screen is wide, though.

The ultrawide aficionados do have a point, even if it’s not really the simple one about “old terminals” they originally thought. Our modern wide-screen displays are criminally underutilized, particularly for text. Even adding in the big chunky file, class, and method tree browser over on the left and the source code preview on the right, a brief survey of a Google Image search for “vs code” shows a lot of editors open with huge, blank areas on the right side of the window.

Big screens are super useful as they allow us to leverage our spatial memories to keep more relevant code around and simply glance around as we think, rather than navigate interactively. But it only works if you remember to do it.

Newspapers allowed us to read a ton of information in one sitting with minimum shuffling by packing in as much as 6 columns of text. You could read a column to the bottom of the page, back to the top, and down again, several times.

Similarly, books fill both of their opposed pages with text at the same time, doubling the amount of stuff you can read at once before needing to turn the page.

You may notice that reading text in a book, even in an ebook app, is more comfortable than reading a ton of text by scrolling around in a web browser. That’s because our eyes are built for saccades, and repeatedly tracking the continuous smooth motion of the page as it scrolls to a stop, then re-targeting the new fixed location to start saccading around from, is literally more physically strenuous on your eye’s muscles!

There’s a reason that the codex was a big technological innovation over the scroll. This is a regression!

Today, the right thing to do here is to make use of horizontally split panes in your text editor or IDE, and just make a bit of conscious effort to set up the appropriate code on screen for the problem you’re working on. However, this is a potential area for different IDEs to really differentiate themselves, and build multi-column continuous-code-reading layouts that allow for buffers to wrap and be navigable newspaper-style.

Similar, modern CSS has shockingly good support for multi-column layouts, and it’s a shame that true multi-column, page-turning layouts are so rare. If I ever figure out a way to deploy this here that isn’t horribly clunky and fighting modern platform conventions like “scrolling horizontally is substantially more annoying and inconsistent than scrolling vertically” maybe I will experiment with such a layout on this blog one day. Until then… just make the browser window narrower so other useful stuff can be in the other parts of the screen, I guess.

Code Isn’t Prose

But, I digress. While I think that columnar layouts for reading prose are an interesting thing more people should experiment with, code isn’t prose.

The metric used for ideal line width, which you may have noticed if you clicked through some of those Wikipedia links earlier, is not “character cells in your editor window”, it is characters per line, or “CPL”.

With an optimal CPL somewhere between 45 and 95, a code-line-width of somewhere around 90 might actually be the best idea, because whitespace uses up your line-width budget. In a typical object-oriented Python program², most of your code ends up indented by at least 8 spaces: 4 for the class scope, 4 for the method scope. Most likely a lot of it is 12, because any interesting code will have at least one conditional or loop. So, by the time you’re done wasting all that horizontal space, a max line length of 90 actually looks more like a maximum of 78... right about that sweet spot from the US-Letter page in the typewriter that we started with.

What about soft-wrap?

In principle, source code is structured information, whose presentation could be fully decoupled from its serialized representation. Everyone could configure their preferred line width appropriate to their custom preferences and the specific physiological characteristics of their eyes, and the code could be formatted according to the language it was expressed in, and “hard wrapping” could be a silly antiquated thing.

The problem with this argument is the same as the argument against “but tabs are semantic indentation”, to wit: nope, no it isn’t. What “in principle” means in the previous paragraph is actually “in a fantasy world which we do not inhabit”. I’d love it if editors treated code this way and we had a rich history and tradition of structured manipulations rather than typing in strings of symbols to construct source code textually. But that is not the world we live in. Hard wrapping is unfortunately necessary to integrate with diff tools.

So what’s the optimal line width?

The exact, specific number here is still ultimately a matter of personal preference.

Hopefully, understanding the long history, science, and underlying physical constraints can lead you to select a contextually appropriate value for your own purposes that will balance ease of reading, integration with the relevant tools in your ecosystem, diff size, presentation in the editors and IDEs that your contributors tend to use, reasonable display in web contexts, on presentation slides, and so on.

But — and this is important — counterpoint:

No it isn’t, you don’t need to select an optimal width, because it’s already been selected for you. It is 88.

Acknowledgments

Thank you for reading, and especially thank you to my patrons who are supporting my writing on this blog. If you like what you’ve read here and you’d like to read more of it, or you’d like to support my various open-source endeavors, you can support my work as a sponsor!

The Problem

Like many other self-styled thinky programmer guys, I like to imagine myself as a sort of Holmesian genius, making trenchant observations, collecting them, and then synergizing them into brilliant deductions with the keen application of my powerful mind.

However, several years ago, I had an epiphany in my self-concept. I finally understood that, to the extent that I am usefully clever, it is less in a Holmesian idiom, and more, shall we say, Monkesque.

For those unfamiliar with either of the respective franchises:

• Holmes is a towering intellect honed by years of training, who catalogues intentional, systematic observations and deduces logical, factual conclusions from those observations.
• Monk, on the other hand, while also a reasonably intelligent guy, is highly neurotic, wracked by unresolved trauma and profound grief. As both a consulting job and a coping mechanism, he makes a habit of erratically wandering into crime scenes, and, driven by a carefully managed jenga tower of mental illnesses, leverages his dual inabilities to solve crimes. First, he is unable to filter out apparently inconsequential details, building up a mental rat’s nest of trivia about the problem; second, he is unable to let go of any minor incongruity, obsessively ruminating on the collection of facts until they all make sense in a consistent timeline.

Perhaps surprisingly, this tendency serves both this fictional wretch of a detective, and myself, reasonably well. I find annoying incongruities in abstractions and I fidget and fiddle with them until I end up building something that a lot of people like, or perhaps something that a smaller number of people get really excited about. At worst, at least I eventually understand what’s going on. This is a self-soothing activity but it turns out that, managed properly, it can very effectively soothe others as well.

All that brings us to today’s topic, which is an incongruity I cannot smooth out or fit into a logical framework to make sense. I am, somewhat reluctantly, a genAI skeptic. However, I am, even more reluctantly, exposed to genAI Discourse every damn minute of every damn day. It is relentless, inescapable, and exhausting.

This preamble about personality should hopefully help you, dear reader, to understand how I usually address problematical ideas by thinking and thinking and fidgeting with them until I manage to write some words — or perhaps a new open source package — that logically orders the ideas around it in a way which allows my brain to calm down and let it go, and how that process is important to me.

In this particular instance, however, genAI has defeated me. I cannot make it make sense, but I need to stop thinking about it anyway. It is too much and I need to give up.

My goal with this post is not to convince anyone of anything in particular — and we’ll get to why that is a bit later — but rather:

1. to set out my current understanding in one place, including all the various negative feelings which are still bothering me, so I can stop repeating it elsewhere,
2. to explain why I cannot build a case that I think should be particularly convincing to anyone else, particularly to someone who actively disagrees with me,
3. in so doing, to illustrate why I think the discourse is so fractious and unresolvable, and finally
4. to give myself, and hopefully by proxy to give others in the same situation, permission to just peace out of this nightmare quagmire corner of the noosphere.

But first, just because I can’t prove that my interlocutors are Wrong On The Internet, doesn’t mean I won’t explain why I feel like they are wrong.

The Anti-Antis

Most recently, at time of writing, there have been a spate of “the genAI discourse is bad” articles, almost exclusively written from the perspective of, not boosters exactly, but pragmatically minded (albeit concerned) genAI users, wishing for the skeptics to be more pointed and accurate in our critiques. This is anti-anti-genAI content.

I am not going to link to any of these, because, as part of their self-fulfilling prophecy about the “genAI discourse”, they’re also all bad.

Mostly, however, they had very little worthwhile to respond to because they were straw-manning their erstwhile interlocutors. They are all getting annoyed at “bad genAI criticism” while failing to engage with — and often failing to even mention — most of the actual substance of any serious genAI criticism. At least, any of the criticism that I’ve personally read.

I understand wanting to avoid a callout or Gish-gallop culture and just express your own ideas. So, I understand that they didn’t link directly to particular sources or go point-by-point on anyone else’s writing. Obviously I get it, since that’s exactly what this post is doing too.

But if you’re going to talk about how bad the genAI conversation is, without even mentioning huge categories of problem like “climate impact” or “disinformation”¹ even once, I honestly don’t know what conversation you’re even talking about. This is peak “make up a guy to get mad at” behavior, which is especially confusing in this circumstance, because there’s an absolutely huge crowd of actual people that you could already be mad at.

The people writing these pieces have historically seemed very thoughtful to me. Some of them I know personally. It is worrying to me that their critical thinking skills appear to have substantially degraded specifically after spending a bunch of time intensely using this technology which I believe has a scary risk of degrading one’s critical thinking skills. Correlation is not causation or whatever, and sure, from a rhetorical perspective this is “post hoc ergo propter hoc” and maybe a little “ad hominem” for good measure, but correlation can still be concerning.

Yet, I cannot effectively respond to these folks, because they are making a practical argument that I cannot, despite my best efforts, find compelling evidence to refute categorically. My experiences of genAI are all extremely bad, but that is barely even anecdata. Their experiences are neutral-to-positive. Little scientific data exists. How to resolve this?²

The Aesthetics

As I begin to state my own position, let me lead with this: my factual analysis of genAI is hopelessly negatively biased. I find the vast majority of the aesthetic properties of genAI to be intensely unpleasant.

I have been trying very hard to correct for this bias, to try to pay attention to the facts and to have a clear-eyed view of these systems’ capabilities. But the feelings are visceral, and the effort to compensate is tiring. It is, in fact, the desire to stop making this particular kind of effort that has me writing up this piece and trying to take an intentional break from the subject, despite its intense relevance.

When I say its “aesthetic qualities” are unpleasant, I don’t just mean the aesthetic elements of output of genAIs themselves. The aesthetic quality of genAI writing, visual design, animation and so on, while mostly atrocious, is also highly variable. There are cherry-picked examples which look… fine. Maybe even good. For years now, there have been, famously, literally award-winning aesthetic outputs of genAI³.

While I am ideologically predisposed to see any “good” genAI art as accruing the benefits of either a survivorship bias from thousands of terrible outputs or simple plagiarism rather than its own inherent quality, I cannot deny that in many cases it is “good”.

However, I am not just talking about the product, but the process; the aesthetic experience of interfacing with the genAI system itself, rather than the aesthetic experience of the outputs of that system.

I am not a visual artist and I am not really a writer⁴, particularly not a writer of fiction or anything else whose experience is primarily aesthetic. So I will speak directly to the experience of software development.

I have seen very few successful examples of using genAI to produce whole, working systems. There are no shortage of highly public miserable failures, particularly from the vendors of these systems themselves, where the outputs are confused, self-contradictory, full of subtle errors and generally unusable. While few studies exist, it sure looks like this is an automated way of producing a Net Negative Productivity Programmer, throwing out chaff to slow down the rest of the team.⁵

Juxtapose this with my aforementioned psychological motivations, to wit, I want to have everything in the computer be orderly and make sense, I’m sure most of you would have no trouble imagining that sitting through this sort of practice would make me extremely unhappy.

Despite this plethora of negative experiences, executives are aggressively mandating the use of AI⁶. It looks like without such mandates, most people will not bother to use such tools, so the executives will need muscular policies to enforce its use.⁷

Being forced to sit and argue with a robot while it struggles and fails to produce a working output, while you have to rewrite the code at the end anyway, is incredibly demoralizing. This is the kind of activity that activates every single major cause of burnout at once.

But, at least in that scenario, the thing ultimately doesn’t work, so there’s a hope that after a very stressful six month pilot program, you can go to management with a pile of meticulously collected evidence, and shut the whole thing down.

I am inclined to believe that, in fact, it doesn’t work well enough to be used this way, and that we are going to see a big crash. But that is not the most aesthetically distressing thing. The most distressing thing is that maybe it does work; if not well enough to actually do the work, at least ambiguously enough to fool the executives long-term.

This project, in particular, stood out to me as an example. Its author, a self-professed “AI skeptic” who “thought LLMs were glorified Markov chain generators that didn’t actually understand code and couldn’t produce anything novel”, did a green-field project to test this hypothesis.

Now, this particular project is not totally inconsistent with a world in which LLMs cannot produce anything novel. One could imagine that, out in the world of open source, perhaps there is enough “OAuth provider written in TypeScript” blended up into the slurry of “borrowed⁸” training data that the minor constraint of “make it work on Cloudflare Workers” is a small tweak⁹. It is not fully dispositive of the question of the viability of “genAI coding”.

But it is a data point related to that question, and thus it did make me contend with what might happen if it were actually a fully demonstrative example. I reviewed the commit history, as the author suggested. For the sake of argument, I tried to ask myself if I would like working this way. Just for clarity on this question, I wanted to suspend judgement about everything else; assuming:

• the model could be created with ethically, legally, voluntarily sourced training data
• its usage involved consent from labor rather than authoritarian mandates
• sensible levels of energy expenditure, with minimal CO2 impact
• it is substantially more efficient to work this way than to just write the code yourself

and so on, and so on… would I like to use this magic robot that could mostly just emit working code for me? Would I use it if it were free, in all senses of the word?

No. I absolutely would not.

I found the experience of reading this commit history and imagining myself using such a tool — without exaggeration — nauseating.

Unlike many programmers, I love code review. I find that it is one of the best parts of the process of programming. I can help people learn, and develop their skills, and learn from them, and appreciate the decisions they made, develop an impression of a fellow programmer’s style. It’s a great way to build a mutual theory of mind.

Of course, it can still be really annoying; people make mistakes, often can’t see things I find obvious, and in particular when you’re reviewing a lot of code from a lot of different people, you often end up having to repeat explanations of the same mistakes. So I can see why many programmers, particularly those more introverted than I am, hate it.

But, ultimately, when I review their code and work hard to provide clear and actionable feedback, people learn and grow and it’s worth that investment in inconvenience.

The process of coding with an “agentic” LLM appears to be the process of carefully distilling all the worst parts of code review, and removing and discarding all of its benefits.

The lazy, dumb, lying robot asshole keeps making the same mistakes over and over again, never improving, never genuinely reacting, always obsequiously pretending to take your feedback on board.

Even when it “does” actually “understand” and manages to load your instructions into its context window, 200K tokens later it will slide cleanly out of its memory and you will have to say it again.

All the while, it is attempting to trick you. It gets most things right, but it consistently makes mistakes in the places that you are least likely to notice. In places where a person wouldn’t make a mistake. Your brain keeps trying to develop a theory of mind to predict its behavior but there’s no mind there, so it always behaves infuriatingly randomly.

I don’t think I am the only one who feels this way.

The Affordances

Whatever our environments afford, we tend to do more of. Whatever they resist, we tend to do less of. So in a world where we were all writing all of our code and emails and blog posts and texts to each other with LLMs, what do they afford that existing tools do not?

As a weirdo who enjoys code review, I also enjoy process engineering. The central question of almost all process engineering is to continuously ask: how shall we shape our tools, to better shape ourselves?

LLMs are an affordance for producing more text, faster. How is that going to shape us?

Again arguing in the alternative here, assuming the text is free from errors and hallucinations and whatever, it’s all correct and fit for purpose, that means it reduces the pain of circumstances where you have to repeat yourself. Less pain! Sounds great; I don’t like pain.

Every codebase has places where you need boilerplate. Every organization has defects in its information architecture that require repetition of certain information rather than a link back to the authoritative source of truth. Often, these problems persist for a very long time, because it is difficult to overcome the institutional inertia required to make real progress rather than going along with the status quo. But this is often where the highest-value projects can be found. Where there’s muck, there’s brass.

The process-engineering function of an LLM, therefore, is to prevent fundamental problems from ever getting fixed, to reward the rapid-fire overwhelm of infrastructure teams with an immediate, catastrophic cascade of legacy code that is now much harder to delete than it is to write.

There is a scene in Game of Thrones where Khal Drogo kills himself. He does so by replacing a stinging, burning, therapeutic antiseptic wound dressing with some cool, soothing mud. The mud felt nice, addressed the immediate pain, removed the discomfort of the antiseptic, and immediately gave him a lethal infection.

The pleasing feeling of immediate progress when one prompts an LLM to solve some problem feels like cool mud on my brain.

The Economics

We are in the middle of a mania around this technology. As I have written about before, I believe the mania will end. There will then be a crash, and a “winter”. But, as I may not have stressed sufficiently, this crash will be the biggest of its kind — so big, that it is arguably not of a kind at all. The level of investment in these technologies is bananas and the possibility that the investors will recoup their investment seems close to zero. Meanwhile, that cost keeps going up, and up, and up.

Others have reported on this in detail¹⁰, and I will not reiterate that all here, but in addition to being a looming and scary industry-wide (if we are lucky; more likely it’s probably “world-wide”) economic threat, it is also going to drive some panicked behavior from management.

Panicky behavior from management stressed that their idea is not panning out is, famously, the cause of much human misery. I expect that even in the “good” scenario, where some profit is ultimately achieved, will still involve mass layoffs rocking the industry, panicked re-hiring, destruction of large amounts of wealth.

It feels bad to think about this.

The Energy Usage

For a long time I believed that the energy impact was overstated. I am even on record, about a year ago, saying I didn’t think the energy usage was a big deal. I think I was wrong about that.

It initially seemed like it was letting regular old data centers off the hook. But recently I have learned that, while the numbers are incomplete because the vendors aren’t sharing information, they’re also extremely bad.¹¹

I think there’s probably a version of this technology that isn’t a climate emergency nightmare, but that’s not the version that the general public has access to today.

The Educational Impact

LLMs are making academic cheating incredibly rampant.¹²

Not only is it so common as to be nearly universal, it’s also extremely harmful to learning.¹³

For learning, genAI is a forklift at the gym.

To some extent, LLMs are simply revealing a structural rot within education and academia that has been building for decades if not centuries. But it was within those inefficiencies and the inconveniences of the academic experience that real learning was, against all odds, still happening in schools.

LLMs produce a frictionless, streamlined process where students can effortlessly glide through the entire credential, learning nothing. Once again, they dull the pain without regard to its cause.

This is not good.

The Invasion of Privacy

This is obviously only a problem with the big cloud models, but then, the big cloud models are the only ones that people actually use. If you are having conversations about anything private with ChatGPT, you are sending all of that private information directly to Sam Altman, to do with as he wishes.

Even if you don’t think he is a particularly bad guy, maybe he won’t even create the privacy nightmare on purpose. Maybe he will be forced to do so as a result of some bizarre kafkaesque accident.¹⁴

Imagine the scenario, for example, where a woman is tracking her cycle and uploading the logs to ChatGPT so she can chat with it about a health concern. Except, surprise, you don’t have to imagine, you can just search for it, as I have personally, organically, seen three separate women on YouTube, at least one of whom lives in Texas, not only do this on camera but recommend doing this to their audiences.

Citation links withheld on this particular claim for hopefully obvious reasons.

I assure you that I am neither particularly interested in menstrual products nor genAI content, and if I am seeing this more than once, it is probably a distressingly large trend.

The Stealing

The training data for LLMs is stolen. I don’t mean like “pirated” in the sense where someone illicitly shares a copy they obtained legitimately; I mean their scrapers are ignoring both norms¹⁵ and laws¹⁶ to obtain copies under false pretenses, destroying other people’s infrastructure¹⁷ in the process.

The Fatigue

I have provided references to numerous articles outlining rhetorical and sometimes data-driven cases for the existence of certain properties and consequences of genAI tools. But I can’t prove any of these properties, either at a point in time or as a durable ongoing problem.

The LLMs themselves are simply too large to model with the usual kind of heuristics one would use to think about software. I’d sooner be able to predict the physics of dice in a casino than a 2 trillion parameter neural network. They resist scientific understanding, not just because of their size and complexity, but because unlike a natural phenomenon (which could of course be considerably larger and more complex) they resist experimentation.

The first form of genAI resistance to experiment is that every discussion is a motte-and-bailey. If I use a free model and get a bad result I’m told it’s because I should have used the paid model. If I get a bad result with ChatGPT I should have used Claude. If I get a bad result with a chatbot I need to start using an agentic tool. If an agentic tool deletes my hard drive by putting os.system(“rm -rf ~/”) into sitecustomize.py then I guess I should have built my own MCP integration with a completely novel heretofore never even considered security sandbox or something?

What configuration, exactly, would let me make a categorical claim about these things? What specific methodological approach should I stick to, to get reliably adequate prompts?

For the record though, if the idea of the free models is that they are going to be provocative demonstrations of the impressive capabilities of the commercial models, and the results are consistently dogshit, I am finding it increasingly hard to care how much better the paid ones are supposed to be, especially since the “better”-ness cannot really be quantified in any meaningful way.

The motte-and-bailey doesn’t stop there though. It’s a war on all fronts. Concerned about energy usage? That’s OK, you can use a local model. Concerned about infringement? That’s okay, somewhere, somebody, maybe, has figured out how to train models consensually¹⁸. Worried about the politics of enriching the richest monsters in the world? Don’t worry, you can always download an “open source” model from Hugging Face. It doesn’t matter that many of these properties are mutually exclusive and attempting to fix one breaks two others; there’s always an answer, the field is so abuzz with so many people trying to pull in so many directions at once that it is legitimately difficult to understand what’s going on.

Even here though, I can see that characterizing everything this way is unfair to a hypothetical sort of person. If there is someone working at one of these thousands of AI companies that have been springing up like toadstools after a rain, and they really are solving one of these extremely difficult problems, how can I handwave that away? We need people working on problems, that’s like, the whole point of having an economy. And I really don’t like shitting on other people’s earnest efforts, so I try not to dismiss whole fields. Given how AI has gotten into everything, in a way that e.g. cryptocurrency never did, painting with that broad a brush inevitably ends up tarring a bunch of stuff that isn’t even really AI at all.

The second form of genAI resistance to experiment is the inherent obfuscation of productization. The models themselves are already complicated enough, but the products that are built around the models are evolving extremely rapidly. ChatGPT is not just a “model”, and with the rapid¹⁹ deployment of Model Context Protocol tools, the edges of all these things will blur even further. Every LLM is now just an enormous unbounded soup of arbitrary software doing arbitrary whatever. How could I possibly get my arms around that to understand it?

The Challenge

I have woefully little experience with these tools.

I’ve tried them out a little bit, and almost every single time the result has been a disaster that has not made me curious to push further. Yet, I keep hearing from all over the industry that I should.

To some extent, I feel like the motte-and-bailey characterization above is fair; if the technology itself can really do real software development, it ought to be able to do it in multiple modalities, and there’s nothing anyone can articulate to me about GPT-4o which puts it in a fundamentally different class than GPT-3.5.

But, also, I consistently hear that the subjective experience of using the premium versions of the tools is actually good, and the free ones are actually bad.

I keep struggling to find ways to try them “the right way”, the way that people I know and otherwise respect claim to be using them, but I haven’t managed to do so in any meaningful way yet.

I do not want to be using the cloud versions of these models with their potentially hideous energy demands; I’d like to use a local model. But there is obviously not a nicely composed way to use local models like this.

Since there are apparently zero models with ethically-sourced training data, and litigation is ongoing²⁰ to determine the legal relationships of training data and outputs, even if I can be comfortable with some level of plagiarism on a project, I don’t feel that I can introduce the existential legal risk into other people’s infrastructure, so I would need to make a new project.

Others have differing opinions of course, including some within my dependency chain, which does worry me, but I still don’t feel like I can freely contribute further to the problem; it’s going to be bad enough to unwind any impact upstream. Even just for my own sake, I don’t want to make it worse.

This especially presents a problem because I have way too much stuff going on already. A new project is not practical.

Finally, even if I did manage to satisfy all of my quirky²¹ constraints, would this experiment really be worth anything? The models and tools that people are raving about are the big, expensive, harmful ones. If I proved to myself yet again that a small model with bad tools was unpleasant to use, I wouldn’t really be addressing my opponents’ views.

I’m stuck.

The Surrender

I am writing this piece to make my peace with giving up on this topic, at least for a while. While I do idly hope that some folks might find bits of it convincing, and perhaps find ways to be more mindful with their own usage of genAI tools, and consider the harm they may be causing, that’s not actually the goal. And that is not the goal because it is just so much goddamn work to prove.

Here, I must return to my philosophical hobbyhorse of sprachspiel. In this case, specifically to use it as an analytical tool, not just to understand what I am trying to say, but what the purpose for my speech is.

The concept of sprachspiel is most frequently deployed to describe the goal of the language game being played, but in game theory, that’s only half the story. Speech — particularly rigorously justified speech — has a cost, as well as a benefit. I can make shit up pretty easily, but if I want to do anything remotely like scientific or academic rigor, that cost can be astronomical. In the case of developing an abstract understanding of LLMs, the cost is just too high.

So what is my goal, then? To be king Canute, standing astride the shore of “tech”, whatever that is, commanding the LLM tide not to rise? This is a multi-trillion dollar juggernaut.

Even the rump, loser, also-ran fragment of it has the power to literally suffocate us in our homes²² if they so choose, completely insulated from any consequence. If the power curve starts there, imagine what the winners in this industry are going to be capable of, irrespective of the technology they’re building - just with the resources they have to hand. Am I going to write a blog post that can rival their propaganda apparatus? Doubtful.

Instead, I will just have to concede that maybe I’m wrong. I don’t have the skill, or the knowledge, or the energy, to demonstrate with any level of rigor that LLMs are generally, in fact, hot garbage. Intellectually, I will have to acknowledge that maybe the boosters are right. Maybe it’ll be OK.

Maybe the carbon emissions aren’t so bad. Maybe everybody is keeping them secret in ways that they don’t for other types of datacenter for perfectly legitimate reasons. Maybe the tools really can write novel and correct code, and with a little more tweaking, it won’t be so difficult to get them to do it. Maybe by the time they become a mandatory condition of access to developer tools, they won’t be miserable.

Sure, I even sincerely agree, intellectual property really has been a pretty bad idea from the beginning. Maybe it’s OK that we’ve made an exception to those rules. The rules were stupid anyway, so what does it matter if we let a few billionaires break them? Really, everybody should be able to break them (although of course, regular people can’t, because we can’t afford the lawyers to fight off the MPAA and RIAA, but that’s a problem with the legal system, not tech).

I come not to praise “AI skepticism”, but to bury it.

Maybe it really is all going to be fine. Perhaps I am simply catastrophizing; I have been known to do that from time to time. I can even sort of believe it, in my head. Still, even after writing all this out, I can’t quite manage to believe it in the pit of my stomach.

Unfortunately, that feeling is not something that you, or I, can argue with.

Acknowledgments

Thank you to my patrons. Normally, I would say, “who are supporting my writing on this blog”, but in the case of this piece, I feel more like I should apologize to them for this than to thank them; these thoughts have been preventing me from thinking more productive, useful things that I actually have relevant skill and expertise in; this felt more like a creative blockage that I just needed to expel than a deliberately written article. If you like what you’ve read here and you’d like to read more of it, well, too bad; I am sincerely determined to stop writing about this topic. But, if you’d like to read more stuff like other things I have written, or you’d like to support my various open-source endeavors, you can support my work as a sponsor!

The History

Before dataclasses were added to Python in version 3.7 — in June of 2018 — the __init__ special method had an important use. If you had a class representing a data structure — for example a 2DCoordinate, with x and y attributes — you would want to be able to construct it as 2DCoordinate(x=1, y=2), which would require you to add an __init__ method with x and y parameters.

The other options available at the time all had pretty bad problems:

1. You could remove 2DCoordinate from your public API and instead expose a make_2d_coordinate function and make it non-importable, but then how would you document your return or parameter types?
2. You could document the x and y attributes and make the user assign each one themselves, but then 2DCoordinate() would return an invalid object.
3. You could default your coordinates to 0 with class attributes, and while that would fix the problem with option 2, this would now require all 2DCoordinate objects to be not just mutable, but mutated at every call site.
4. You could fix the problems with option 1 by adding a new abstract class that you could expose in your public API, but this would explode the complexity of every new public class, no matter how simple. To make matters worse, typing.Protocol didn’t even arrive until Python 3.8, so, in the pre-3.7 world this would condemn you to using concrete inheritance and declaring multiple classes even for the most basic data structure imaginable.

Also, an __init__ method that does nothing but assign a few attributes doesn’t have any significant problems, so it is an obvious choice in this case. Given all the problems that I just described with the alternatives, it makes sense that it became the obvious default choice, in most cases.

However, by accepting “define a custom __init__” as the default way to allow users to create your objects, we make a habit of beginning every class with a pile of arbitrary code that gets executed every time it is instantiated.

Wherever there is arbitrary code, there are arbitrary problems.

The Problems

Let’s consider a data structure more complex than one that simply holds a couple of attributes. We will create one that represents a reference to some I/O in the external world: a FileReader.

Of course Python has its own open-file object abstraction, but I will be ignoring that for the purposes of the example.

Let’s assume a world where we have the following functions, in an imaginary fileio module:

• open(path: str) -> int
• read(fileno: int, length: int)
• close(fileno: int)

Our hypothetical fileio.open returns an integer representing a file descriptor¹, fileio.read allows us to read length bytes from an open file descriptor, and fileio.close closes that file descriptor, invalidating it for future use.

With the habit that we have built from writing thousands of __init__ methods, we might want to write our FileReader class like this:

1
2
3
4
5
6
7

class FileReader:
    def __init__(self, path: str) -> None:
        self._fd = fileio.open(path)
    def read(self, length: int) -> bytes:
        return fileio.read(self._fd, length)
    def close(self) -> None:
        fileio.close(self._fd)

For our initial use-case, this is fine. Client code creates a FileReader by doing something like FileReader("./config.json"), which always creates a FileReader that maintains its file descriptor int internally as private state. This is as it should be; we don’t want user code to see or mess with _fd, as that might violate FileReader’s invariants. All the necessary work to construct a valid FileReader — i.e. the call to open — is always taken care of for you by FileReader.__init__.

However, additional requirements will creep in, and as they do, FileReader.__init__ becomes increasingly awkward.

Initially we only care about fileio.open, but later, we may have to deal with a library that has its own reasons for managing the call to fileio.open by itself, and wants to give us an int that we use as our _fd, we now have to resort to weird workarounds like:

1
2
3
4

def reader_from_fd(fd: int) -> FileReader:
    fr = object.__new__(FileReader)
    fr._fd = fd
    return fr

Now, all those nice properties that we got from trying to force object construction to give us a valid object are gone. reader_from_fd’s type signature, which takes a plain int, has no way of even suggesting to client code how to ensure that it has passed in the right kind of int.

Testing is much more of a hassle, because we have to patch in our own copy of fileio.open any time we want an instance of a FileReader in a test without doing any real-life file I/O, even if we could (for example) share a single file descriptor among many FileReader s for testing purposes.

All of this also assumes a fileio.open that is synchronous. Although for literal file I/O this is more of a hypothetical concern, there are many types of networked resource which are really only available via an asynchronous (and thus: potentially slow, potentially error-prone) API. If you’ve ever found yourself wanting to type async def __init__(self): ... then you have seen this limitation in practice.

Comprehensively describing all the possible problems with this approach would end up being a book-length treatise on a philosophy of object oriented design, so I will sum up by saying that the cause of all these problems is the same: we are inextricably linking the act of creating a data structure with whatever side-effects are most often associated with that data structure. If they are “often” associated with it, then by definition they are not “always” associated with it, and all the cases where they aren’t associated become unweildy and potentially broken.

Defining an __init__ is an anti-pattern, and we need a replacement for it.

The Solutions

I believe this tripartite assemblage of design techniques will address the problems raised above:

• using dataclass to define attributes,
• replacing behavior that previously would have previously been in __init__ with a new classmethod that does the same thing, and
• using precise types to describe what a valid instance looks like.

Using dataclass attributes to create an __init__ for you

To begin, let’s refactor FileReader into a dataclass. This does get us an __init__ method, but it won’t be one an arbitrary one we define ourselves; it will get the useful constraint enforced on it that it will just assign attributes.

1
2
3
4
5
6
7

@dataclass
class FileReader:
    _fd: int
    def read(self, length: int) -> bytes:
        return fileio.read(self._fd, length)
    def close(self) -> None:
        fileio.close(self._fd)

Except... oops. In fixing the problems that we created with our custom __init__ that calls fileio.open, we have re-introduced several problems that it solved:

1. We have removed all the convenience of FileReader("path"). Now the user needs to import the low-level fileio.open again, making the most common type of construction both more verbose and less discoverable; if we want users to know how to build a FileReader in a practical scenario, we will have to add something in our documentation to point at a separate module entirely.
2. There’s no enforcement of the validity of _fd as a file descriptor; it’s just some integer, which the user could easily pass an incorrect instance of, with no error.

In isolation, dataclass by itself can’t solve all our problems, so let’s add in the second technique.

Using classmethod factories to create objects

We don’t want to require any additional imports, or require users to go looking at any other modules — or indeed anything other than FileReader itself — to figure out how to create a FileReader for its intended usage.

Luckily we have a tool that can easily address all of these concerns at once: @classmethod. Let’s define a FileReader.open class method:

1
2
3
4
5
6
7

from typing import Self
@dataclass
class FileReader:
    _fd: int
    @classmethod
    def open(cls, path: str) -> Self:
        return cls(fileio.open(path))

Now, your callers can replace FileReader("path") with FileReader.open("path"), and get all the same benefits.

Additionally, if we needed to await fileio.open(...), and thus we needed its signature to be @classmethod async def open, we are freed from the constraint of __init__ as a special method. There is nothing that would prevent a @classmethod from being async, or indeed, from having any other modification to its return value, such as returning a tuple of related values rather than just the object being constructed.

Using NewType to address object validity

Next, let’s address the slightly trickier issue of enforcing object validity.

Our type signature calls this thing an int, and indeed, that is unfortunately what the lower-level fileio.open gives us, and that’s beyond our control. But for our own purposes, we can be more precise in our definitions, using NewType:

1
2

from typing import NewType
FileDescriptor = NewType("FileDescriptor", int)

There are a few different ways to address the underlying library, but for the sake of brevity and to illustrate that this can be done with zero run-time overhead, let’s just insist to Mypy that we have versions of fileio.open, fileio.read, and fileio.write which actually already take FileDescriptor integers rather than regular ones.

1
2
3
4

from typing import Callable
_open: Callable[[str], FileDescriptor] = fileio.open  # type:ignore[assignment]
_read: Callable[[FileDescriptor, int], bytes] = fileio.read
_close: Callable[[FileDescriptor], None] = fileio.close

We do of course have to slightly adjust FileReader, too, but the changes are very small. Putting it all together, we get:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

from typing import Self
@dataclass
class FileReader:
    _fd: FileDescriptor
    @classmethod
    def open(cls, path: str) -> Self:
        return cls(_open(path))
    def read(self, length: int) -> bytes:
        return _read(self._fd, length)
    def close(self) -> None:
        _close(self._fd)

Note that the main technique here is not necessarily using NewType specifically, but rather aligning an instance’s property of “has all attributes set” as closely as possible with an instance’s property of “fully valid instance of its class”; NewType is just a handy tool to enforce any necessary constraints on the places where you need to use a primitive type like int, str or bytes.

In Summary - The New Best Practice

From now on, when you’re defining a new Python class:

• Make it a dataclass².
• Use its default __init__ method³.
• Add @classmethods to provide your users convenient and discoverable ways to build your objects.
• Require that all dependencies be satisfied by attributes, so you always start with a valid object.
• Use typing.NewType to enforce any constraints on primitive data types (like int and str) which might have magical external attributes, like needing to come from a particular library, needing to be random, and so on.

If you define all your classes this way, you will get all the benefits of a custom __init__ method:

• All consumers of your data structures will receive valid objects, because an object with all its attributes populated correctly is inherently valid.
• Users of your library will be presented with convenient ways to create your objects that do as much work as is necessary to make them easy to use, and they can discover these just by looking at the methods on your class itself.

Along with some nice new benefits:

• You will be future-proofed against new requirements for different ways that users may need to construct your object.
• If there are already multiple ways to instantiate your class, you can now give each of them a meaningful name; no need to have monstrosities like def __init__(self, maybe_a_filename: int | str | None = None):
• Your test suite can always construct an object by satisfying all its dependencies; no need to monkey-patch anything when you can always call the type and never do any I/O or generate any side effects.

Before dataclasses, it was always a bit weird that such a basic feature of the Python language — giving data to a data structure to make it valid — required overriding a method with 4 underscores in its name. __init__ stuck out like a sore thumb. Other such methods like __add__ or even __repr__ were inherently customizing esoteric attributes of classes.

For many years now, that historical language wart has been resolved. @dataclass, @classmethod, and NewType give you everything you need to build classes which are convenient, idiomatic, flexible, testable, and robust.

Acknowledgments

Thank you to my patrons who are supporting my writing on this blog. If you like what you’ve read here and you’d like to read more of it, or you’d like to support my various open-source endeavors, you can support my work as a sponsor! I am also available for consulting work if you think your organization could benefit from expertise on topics like “but what is a ‘class’, really?”.

A Database File

When I was 10 years old, and going through a fairly difficult time, I was lucky enough to come into the possession of a piece of software called Claris FileMaker Pro™.

FileMaker allowed its users to construct arbitrary databases, and to associate their tables with a customized visual presentation. FileMaker also had a rudimentary scripting language, which would allow users to imbue these databases with behavior.

As a mentally ill pre-teen, lacking a sense of control over anything or anyone in my own life, including myself, I began building a personalized database to catalogue the various objects and people in my immediate vicinity. If one were inclined to be generous, one might assess this behavior and say I was systematically taxonomizing the objects in my life and recording schematized information about them.

As I saw it at the time, if I collected the information, I could always use it later, to answer questions that I might have. If I didn’t collect it, then what if I needed it? Surely I would regret it! Thus I developed a categorical imperative to spend as much of my time as possible collecting and entering data about everything that I could reasonably arrange into a common schema.

Having thus summoned this specter of regret for all lost data-entry opportunities, it was hard to dismiss. We might label it “Claris’s Basilisk”, for obvious reasons.

Therefore, a less-generous (or more clinically-minded) observer might have replaced the word “systematically” with “obsessively” in the assessment above.

I also began writing what scripts were within my marginal programming abilities at the time, just because I could: things like computing the sum of every street number of every person in my address book. Why was this useful? Wrong question: the right question is “was it possible” to which my answer was “yes”.

If I was obliged to collect all the information which I could observe — in case it later became interesting — I was similarly obliged to write and run every program I could. It might, after all, emit some other interesting information.

I was an avid reader of science fiction as well.

I had this vague sense that computers could kind of think. This resulted in a chain of reasoning that went something like this:

1. human brains are kinda like computers,
2. the software running in the human brain is very complex,
3. I could only write simple computer programs, but,
4. when you really think about it, a “complex” program is just a collection of simpler programs

Therefore: if I just kept collecting data, collecting smaller programs that could solve specific problems, and connecting them all together in one big file, eventually the database as a whole would become self-aware and could solve whatever problem I wanted. I just needed to be patient; to “keep grinding” as the kids would put it today.

I still feel like this is an understandable way to think — if you are a highly depressed and anxious 10-year-old in 1990.

Anyway.

35 Years Later

OpenAI is a company that produces transformer architecture machine learning generative AI models; their current generation was trained on about 10 trillion words, obtained in a variety of different ways from a large variety of different, unrelated sources.

A few days ago, on March 26, 2025 at 8:41 AM Pacific Time, Sam Altman took to “X™, The Everything App™,” and described the trajectory of his career of the last decade at OpenAI as, and I quote, a “grind for a decade trying to help make super-intelligence to cure cancer or whatever” (emphasis mine).

I really, really don’t want to become a full-time AI skeptic, and I am not an expert here, but I feel like I can identify a logically flawed premise when I see one.

This is not a system-design strategy. It is a trauma response.

You can’t cure cancer “or whatever”. If you want to build a computer system that does some thing, you actually need to hire experts in that thing, and have them work to both design and validate that the system is fit for the purpose of that thing.

Aside: But... are they, though?

I am not an oncologist; I do not particularly want to be writing about the specifics here, but, if I am going to make a claim like “you can’t cure cancer this way” I need to back it up.

My first argument — and possibly my strongest — is that cancer is not cured.

QED.

But I guess, to Sam’s credit, there is at least one other company partnering with OpenAI to do things that are specifically related to cancer. However, that company is still in a self-described “initial phase” and it’s not entirely clear that it is going to work out very well.

Almost everything I can find about it online was from a PR push in the middle of last year, so it all reads like a press release. I can’t easily find any independently-verified information.

A lot of AI hype is like this. A promising demo is delivered; claims are made that surely if the technology can solve this small part of the problem now, within 5 years surely it will be able to solve everything else as well!

But even the light-on-content puff-pieces tend to hedge quite a lot. For example, as the Wall Street Journal quoted one of the users initially testing it (emphasis mine):

The most promising use of AI in healthcare right now is automating “mundane” tasks like paperwork and physician note-taking, he said. The tendency for AI models to “hallucinate” and contain bias presents serious risks for using AI to replace doctors. Both Color’s Laraki and OpenAI’s Lightcap are adamant that doctors be involved in any clinical decisions.

I would probably not personally characterize “‘mundane’ tasks like paperwork and … note-taking” as “curing cancer”. Maybe an oncologist could use some code I developed too; even if it helped them, I wouldn’t be stealing valor from them on the curing-cancer part of their job.

Even fully giving it the benefit of the doubt that it works great, and improves patient outcomes significantly, this is medical back-office software. It is not super-intelligence.

It would not even matter if it were “super-intelligence”, whatever that means, because “intelligence” is not how you do medical care or medical research. It’s called “lab work” not “lab think”.

To put a fine point on it: biomedical research fundamentally cannot be done entirely by reading papers or processing existing information. It cannot even be done by testing drugs in computer simulations.

Biological systems are enormously complex, and medical research on new therapies inherently requires careful, repeated empirical testing to validate the correspondence of existing research with reality. Not “an experiment”, but a series of coordinated experiments that all test the same theoretical model. The data (which, in an LLM context, is “training data”) might just be wrong; it may not reflect reality, and the only way to tell is to continuously verify it against reality.

Previous observations can be tainted by methodological errors, by data fraud, and by operational mistakes by practitioners. If there were a way to do verifiable development of new disease therapies without the extremely expensive ladder going from cell cultures to animal models to human trials, we would already be doing it, and “AI” would just be an improvement to efficiency of that process. But there is no way to do that and nothing about the technologies involved in LLMs is going to change that fact.

Knowing Things

The practice of science — indeed any practice of the collection of meaningful information — must be done by intentionally and carefully selecting inclusion criteria, methodically and repeatedly curating our data, building a model that operates according to rules we understand and can verify, and verifying the data itself with repeated tests against nature. We cannot just hoover up whatever information happens to be conveniently available with no human intervention and hope it resolves to a correct model of reality by accident. We need to look where the keys are, not where the light is.

Piling up more and more information in a haphazard and increasingly precarious pile will not allow us to climb to the top of that pile, all the way to heaven, so that we can attack and dethrone God.

Eventually, we’ll just run out of disk space, and then lose the database file when the family gets a new computer anyway.

Acknowledgments

Thank you to my patrons who are supporting my writing on this blog. If you like what you’ve read here and you’d like to read more of it, or you’d like to support my various open-source endeavors, you can support my work as a sponsor! Special thanks also to Itamar Turner-Trauring and Thomas Grainger for pre-publication feedback on this article; any errors of course remain my own.