Now, if the debate is really about which option is more cost effective, then we could easily run an A/B test to find out. Though TBH my instinct is that that experiment is likely to cost more than the potential cost savings.

What I will say is that my own sense from experimenting around in a non-rigorous way is that the answer depends on how you use the tool. For actual vibecoding you should always go for the SOTA model because it will need less oversight. It’s also less likely to get stuck in a vicious loop that fruitlessly wastes tokens. But for a more hands-on approach where you move in small, carefully planned increments that you review and test in human-comprehensible chunks, smaller models may be preferable. SOTA ones don’t do that much better when working that way, and the slower inference adds a detrimental amount of friction to the work cycle.

At my last employer we were contending with real customer fatigue and attrition. The application was changing so rapidly that it was overwhelming both our ability to uphold quality standards, and customers' ability to keep up with the feature churn.

Or, to put it more succinctly: nobody actually wants all this slop.

There's also a part of me that misses the earlier days of Hacker News when the subject matter was much more eclectic and seemingly every day I'd have a chance to learn about something completely different.

And that's maybe what I miss. Nowadays Hacker News feels increasingly like another social media platform: mostly a lot of opinions, war stories, and self promotion. That feeling of serendipity that so characterized earlier ages of the Web is fading here, too.

Meaning that they didn’t just fail to demonstrate that these supplements could lead to superhuman performance. They didn’t even just fail to demonstrate that they enhance performance. They failed to demonstrate that their overall effect isn’t deleterious.

And yet everyone involved believed quite firmly, based in part on personal experience, that it was self-evident that these supplements would make them better athletes.

A lot of the arguments for AI in education aren’t far off from the reasoning behind the kinds of popular supplements they used in the Enhanced Games. The limits of that style of thinking are demonstrated quite nicely in the 2010 Canadian Medical Association Journal article “Cigarette smoking: an underused tool in high-performance endurance training”: https://pmc.ncbi.nlm.nih.gov/articles/PMC3001541/

My concern is that AI-generated translations and example text still tend to get the subtleties wrong. I’ve seen that cause a lot of confusion for beginning learners who don’t know their target language well enough to know when something isn’t quite right. The help forum on a learner community I’m in is full of cases where the answer is, “You’re confused because what the AI told you is incorrect.”

And that’s a big part of why I still prefer a good old fashioned keyword search over GenAI when I’m looking for more examples to help me understand a new word.

Before we get ahead of ourselves and hype up AI as a tool for making people smarter, let’s first demonstrate that AI really does help them learn better. Otherwise we risk wasting a lot of time and money on silicon snake oil.

Zoom in, though, and it also impoverished my hometown and many others like it. A whole generation saw their job prospects gutted, and had no realistic options for a career shift because it’s really hard to get an entry level position when you’re middle aged; doubly so when there’s a glut of middle aged people looking for jobs.

And waving away the pain those people experience with an appeal to large scale, intergenerational, historical trends is simply heartless.

Which, ironically, is something I’d love for my browser’s JavaScript runtime to do. But I can also respect others not wanting that.

Which, from what I’ve experienced so far, does seem to take a whole lot more effort if you’re using a coding agent. I spend an incredible amount of time making sure mine doesn’t bloat our codebase with Java-flavored Python, and all the noise and defects and performance problems that it brings.

From a Chinese perspective I expect China’s largest customer is China. These days it’s actually kind of wild how many Chinese consumer products aren’t (and won’t be) available at US retailers. And a lot of them are quite good.

For a while now China’s wanted to reduce its dependence on the US for a variety of reasons. And undermining the US tech industry, whose products the US government likes to use as a cudgel, may serve that goal quite nicely.

The challenge for these “Zen and my job” types of work is that much of Zen practice is (deliberately) a bit of an inkblot test. If there’s any real Big Truth to it, it’s that it’s all one big muddle that everyone has to muddle through for themself. Unlike in many Western religious traditions, there is no presumption of univocality - even within the canonical literature, every opinion is understood to be the author’s alone and represents their own personal attempts at muddling things through. It’s just that some people and works are regarded as being particularly worth a read.

But that’s never explicitly stated. Which can be particularly confusing to people who grew up in religious traditions that do presume univocality and a greater degree of spiritual authority. Hence these works by people who see one particular bit that clicks for them and conclude, “Ah, this is what Zen is really about, I must go tell others!” without realizing that perhaps they just stumbled across an inkblot that they found to be particularly evocative.

This is nothing new. For about 30 years now Microsoft has been constantly repeating various flavors of this “make it so a thing can automatically and silently run programs as soon as it touches your computer” thing. It’s always done in the name of user convenience. It always ends up being a fiasco. I don’t know why they keep doing it, it’s not like the exact same PHB keeps making the same decision over and over for 30 years. It’s probably one or a combination of the many well documented flavors of stupid that are deeply baked into the company’s organizational culture.

(And before the inevitable response, no this is not defending Microsoft. Pointing out that an organization’s culture is too deeply, chronically stupid to avoid opening the exact same obvious and gaping security hole over and over and over and over again is not the same as saying, “it’s fine, actually.”)