Boring is good
jenson.org
jenson.org
Anyway, boring is bad. Boring is what spends your attention on irrelevant things. Cobol's syntax is boring in a bad way. Go's error handling is boring in a bad way. Manually clicking through screens again and again because you failed to write UI tests is boring in a bad way.
What could be "boring in a good way" is something that gets things done and gets out of your way. Things like HTTPS, or S3, or your keyboard once you have leaned touch typing, are "boring in a good way". They have no concealed surprises, are well-tested in practice, and do what they say on the tin, every time.
New and shiny things can be "boring in the good way", e.g. uv [2]. Old and established things can be full of (nasty) surprises, and, in this regard, the opposite of boring, e.g. C++.
Cobol was (and for some, still is) exciting at first, but _becomes_ boring once you master it, and the ecosystem evolves to fix or work around its shortcomings. Believe it or not, even UX/UI testers can deal with and find happiness in clicking through UIs for the tenth thousand time (sure, last time I saw such Tester, was at around 2010).
This doesn't mean the technology itself becomes bad or stays good. It just means the understanding (and usage patterns) solidifies, so it becomes less exciting, hence: "boring".
But you can't sell a book with the title "Choose well-established technology". Because people would be like, no sht, Sherlock, I don't need a book to know that.
All this was my understanding before, so not sure why you think "boring" was meant to be equivalent to "old"
As much as I do get the idea, I can see how promoting use of tedious to use and dull tools is something that really misses the mark.
Well known and mature tools are still sharp and lots of them are not tedious to use.
I do not want my browser to be exiting. I do not want for it to change every week. Say moving buttons to different places. Changing how address bar operates. Maybe trying new short cut keys...
Same goes for most useful software. I actually do want them to be dull. And do their job and not get in between and make my day more interesting by having to fight against it.
I picked "sharp" not "exciting".
Dull knife doesn't do its job, you want tools that do the job efficiently. "Boring" as it seems was interpreted as picking tools that don't do the job efficiently. That is why original idea creator found "Choose boring technology" most likely misunderstood.
Boring is good. I don't want to be excited by technology, i want to be bored in the sense that it's simple and out of my way.
Same for KISS. I tend to tell people to not only keep things simple, but boring even. Some new code i need to read and fix or extend? I want to be bored. Bored means it's obvious and simple.
The difference? There are many complex libraries. By definition they are not simple technology.
For example a crypto library. Probably one of the most complex tasks. I would consider it a good library if it's boring to use/extend/read.
Well, yes, but only in the sense that people kept giving him beef about how boring is a bad word in their mind, not because it was a bad word for this context per se. Which is somewhat ironic given your comment!
I suppose what you're getting at is the difference between boring, and "boooooriiiiiing".
Everything at some point will be interpreted incorrectly.
In reality, a choice of word that is interpreted close to correctly by most people is a better choice of words than a choice of word that is interpreted close to incorrectly by most people.
I interpreted it as implying all communication is imperfect, more or less. Each speaker and listener uses language from their own personal context, and reads meaning into each word from their experience, education, even personal preferences.
As for "boring", I feel it was the right word at the time. It's a rhetorical trick, maybe a click bait, to use a word in an unusual way to get attention. Here the word boring, typically negative, is used in a positive encouraging manner. That makes me want to read more to find out why.
An optimum might exist, but it might not be meaningfully better than any other option.
Go's error handling is useful if your system needs to be correct more than it needs to be available, and you should be manually clicking through screens even if you wrote UI tests.
Looking at why something is boring and who it's boring for can be helpful, especially in a team setting. It's entirely possible to automate away a bottleneck that exists for a good reason while sitting across from someone who'd love to handle the issue correctly.
If an LLM can solve a complex problem 50% of the time, then that is still very valuable. But if you are writing a system of small LLMs doing small tasks, then even 1% error rates can compound into highly unreliable systems when stacked together.
The cost of LLMs occasionally giving you wrong answers is worth it for answers to harder tasks, in a way that it is not worth it for smaller tasks. For those smaller tasks, usually you can get much closer to 100% reliability, and more importantly much greater predictability, with hand-engineered code. This makes it much harder to find areas where small LLMs can add value for small boring tasks. Better auto-complete is the only real-world example I can think of.
I'd adjust that statement - If an LLM can solve a complex problem 50% of the time and I can evaluate correctness of the output, then that is still very valuable. I've seen too many people blindly pass on LLM output - for a short while it was a trend in the scientific literature to have LLMs evaluate output of other LLMs? Who knows how correct that was. Luckily that has ended.
That said, I do think there are lots of problems where verification is easier than doing the task itself, especially in computer science. I think it is easier to list tasks that aren't easier to verify than to do from scratch actually. Security is one major one.
I misread this the first time and realised both interpretations are happening. I've seen people copy-paste out of ChatGPT without reading, and I've seen people "pass on" or reject content simply because it has been AI generated.
Highly reliable. So much so that is basically how modern LLMs work internally. Also speaking from personal experience in the projects I work on, it is the chief way to counteract hallucination, poisoned context windows, and scaling beyond the interaction limit.
LLMs evaluating LLM output works surprisingly well.
What do you mean it ended? I still see tons of NLP papers with this methodology.
The emperor's new clothes ...
take your non-existent 2018 "paper" and shove it up your 2025 ass, you troll, if you even have an ass, that is.
and see my critical appraisal of your so-called "critical appraisal", just above.
check out that rudyard kipling poem (or some such racist colonial brit dick), where he says ... something something ... and you'll be a man, my son.
you'll never be a man, troll. you don't have the gumption to ever be one.
gesundheit! bless you, coz u really need it ... boy.
If he means they will never outperform humans at cognitive or robotics tasks, that's a strong claim!
If he just means they aren't conscious... then let's don't debate it any more here. :-)
I agree that we could be in a bubble at the moment though.
- LLM's are too limited in capabilities and make too many mistakes - We're still in the DOS era of LLM's
I'm leaning more towards the the 2nd, but in either case pandora's box has been opened and you can already see the effects of the direction our civilization is moving towards with this technology.
Many businesses have made bad judgements of where the distinction is, some don't even recognise a distinction. This will improve over time.
Small models doing interesting (boring to the author) use-cases is a fine frontier!
I don't agree at all with this though:
> "LLMs are not intelligent and they never will be."
LLMs already write code better than most humans. The problem is we expect them to one-shot things that a human may spend many hours/days/weeks/months doing. We're lacking coordination for long-term LLM work. The models themselves are probably even more powerful than we realize, we just need to get them to "think" as long as a human would.
If you mean better than most humans considering the set of all humans, sure. But they write code worse than most humans who have learned how to write code. That's not very promising for them developing intelligence.
I have fully integrated LLMs into May daily workflow, and am routinely surprised at the innovative solutions that Claude sometimes comes up with. There is a lot of generated code that is far beyond my own capabilities, even as an expert in the field in which I’m working.
In human terms, we would call that knowing how to bullshit. But just like a college student hitting junior year, sooner or later you'll learn that bullshitting only gets you so far.
That's what we've really done. We've taught computers how to bullshit. We've also managed to finally invent something that lets us communicate relatively directly with a computer using human languages. The language processing capabilities of an LLM are an astonishing multi-generational leap. These types of models will absolutely be the foundation for computing interfaces in the future. But they're still language models.
To me it feels like we've invented a new keyboard, and people are fascinated by the stories the thing produces.
Image generation is a bit better, except it's still not really aware of what the picture is, either. It's aware of what images are described as by others, let alone the truth of the generated image. It makes pictures of dragons quite well, but if you ask it for a contour map of a region, is it going to represent it accurately? It's not concerned about truth, it's concerned about truthiness or the appearance of truth. We know when that distinction is important. It doesn't.
Transformer architectures are not replicas of human brain architrcture, but they are not categorically different either.
For what it's worth, this is not some kind of slam dunk fundamental permanent limitation of AI that is constructed from LLMs. Multi-modal learning gets you part of the way. Tool use gets you more of the way, enabling interaction with at least something. Embodiment and autonomy would get you further, but at some point, you need a true online learner that can update its own weights and not just simulate that with a very large context window.
Whether or not this entails any limitation in capability (as in, there is anything a human or animal can actually do cognitively that a LLM AI can't) is an open question, but it is a real difference. The person you're responding to, no matter how similar their activation loops may be to software, didn't develop all behaviors and predictive modeling they currently do by reading and then had their brain frozen in time.
They always seem to miss another metric, saved human brain power. My experience hasn’t been that I can complete tasks faster from go to woe, it’s been that they’re far less mentally taxing and I can do other things whilst the ai churns some of the tedium away. When I’m feeling real tired, it’s a lot easier to review ai code than write my own from scratch. As any writer will tell you, editing is easier than writing from a blank page.
A good analogy might be that I can vacuum the house way faster than my robot vacuum can, probably in less than 1/3rd of the time! But letting the robot do it is still way better for quality of life and also it frees me up to do other things. This has been the true gain from using coding LLM’s in my lived experience.
I find I struggle with this. If an LLM is being used to improve the overall quality of my writing, I feel like it is then doing the writing and that doesn't sit well with me. Same with having it write my code (though I don't write code for a living).
Maybe it would be similar to playing a guitar solo and then having an LLM fix all the missed notes. Is that still my solo? I tend to feel it isn't.
Just my personal struggle with this new, admittedly incredible, paradigm.
I have nothing against cloud or AI per se, but I still believe in the right tool for the right job and in not doing things just for the sake of it. While raising valuation is a good thing, raising costs, delaying more useful features and adding complexity should also be taken into account.
> We keep asking them to do “intelligent things” and find out a) they really aren’t that good at it, and b) replacing that human task is far more complex than we originally thought
I never thought that. From the beginning on this were easy to uncover marketing statements.
um, dynamo is a generator, it takes mechanical energy and turns into to electricity.
Shots fired
Yes I know, stable things are stable
because "for every 1% unemployment goes up 40,000 people die." bubbles bursting hurt people and there's tens of billions in this bubble.
That's just your experience, based on your geolocation and chain of events.
One can argue that every other field of engineering outside of Software Engineering, specializes in making complex things into boring things.
We are the unique snowflakes that take business use cases and build castle in the clouds that may or may not actually solve the business problem at hand.
... and if it all falls down, don't blame us - you clicked the EULA /s
The writer fails to grasp what rabbit hole we've gone down since the 70s/90s when we began applying the principles of prediction to computation, then horizontalized them via web. This was the most exciting time because it added a vast illusory value to the arbitrary, it's a time of massive piracy that posed as corporatocracy.
Once this prediction became automated by AI, yes, now the piracy becomes boring, and in turn reveals what was going on all along.