tablatom

"If we wish to count lines of code, we should not regard them as lines produced but as lines spent." — Dijkstra

When GPT gives an answer like that, is it actually a meaningful description of its capabilities? Does it have that kind of self-awareness? Or is it just a plausible answer based on the training corpus?

Genuine question.

> as art it's pretty worthless

My experience browsing midjourney supports this. It's amazing how quickly you go from "wow" to "whatever". But that's probably cold comfort for a huge number of jobbing illustrators. Paying clients typically don't want art.

Is GPT3 used to process the data itself? As opposed to something like automatically generating validation rules. If yes, how does this avoid creating plausible but incorrect data? e.g. the blurb mentions "It is important to identify and fill in missing data as accurately as possible", which sounds like a fantasy.

I feel this underestimates the seriousness of the difficulties we are facing in the area of social cohesion. The conflating of real/non-real and credible/non-credible is very much at the heart of the Trump/Brexit divide

> learn everything the knowledge engine knows

Isn't this an effectively infinite set? Wolfram Alpha could be said to know "all the numbers", and "all the formulas".

> LLMs can learn structured domains too if they are well-represented in the training set

But can they learn how to apply structured knowledge in precise ways? In mathematical or computational ways? I don't follow the field in great detail but the commentary I read seems to be saying this is not at all the case. And my own experiments with ChatGPT show it has no systematic grasp of logic.

That's pretty radical on the diet. Was it difficult to adjust? I'm pretty thin as it is so I'd be worried I'd waste away!

Is the Oura a good purchase? I've been really eyeing the health benefits of the Apple watch but there's no way I want a portal into the digital world that's a flick of the wrist away. It's bad enough with the phone pickups.

This is turning into a great thread and this comment is the most impactful one for me. I'm going to give it a go. My daughter is 18 and at least half her peer group are in a dreadful state in terms of mental health. You have to wonder if the impact of the modern media environment is way worse than we've comprehended. Of course that's totally unempirical but is the funding there for the studies we need? In any case it's easy to run the personal experiment, as you've suggested.

I'll second this. Based on my own experience (I'm 50) it's all about energy levels. When my energy is high, I can code like I did two decades ago. Probably better. I guess the difference is that in your twenties, energy is almost for free. As you get older, you've got to be smarter about it - all the usual things - diet, exercise, avoiding stress...

Not that I'm good about those things - I do OK but not great. Sometimes (rarely, thankfully) my energy is so low I can't really code at all, sometimes I'm flying. I'd say by far the biggest factor is my overall state of positivity and optimism. And yeah, life definitely gets more complicated as you get older.

Can you provide any details about the harder problems you mention?

The understanding of the rules is very impressive, as is the solution to the first, trivial problem, but things quickly go sideways when the problem gets a bit more complex. I tried to coax it towards the solution, but it didn't help much (I guess I could have persisted)

I am somewhat (but not entirely!) less concerned about becoming redundant after this test.

Share a link to your project?

Nhost is a very similar offering, built on Hasura

https://nhost.io/

The first review on that page[1] is a well argued counterpoint to Graham's essay. Getting your ideas understood is often only one part of what you're trying to achieve.

If that really is your only goal, then I agree, the simpler the better.

[1] https://www.goodreads.com/review/show/755379555?book_show_ac...

I was thinking the same thing. If Google start logging license plate numbers (and, hey, it's Google), they could build a database of known dangerous vehicles and automatically give them a wide berth.

You'd want to get smarter, not dumber: taking into account the fact that the unpredictable (to the car behind) braking could in itself cause an accident, and calculating a best compromise to avoid both collisions, with priority to the more dangerous of the two (the fender-bender could be a great option if you avoid another car doing 50). I'd be surprised if their software doesn't support this kind of thing already.

So glad to see this book mentioned. The follow-up, Turning Pro, which is effectively the second half of the same book, is also golden.

It is so important that something like this happens. The mental image of all the knowledge workers of the world riding tricycles because a real bike is "too difficult", is absolutely perfect.

My understanding of this article was that he pulled off the hack entirely by himself, without having to get someone to visit his page. Maybe I misunderstood.

> Custom JS on your subdomains is a bad idea

What is the difference between allowing users to put custom JS on a subdomain, vs. someone just opening up the developer console and running whatever JS they like? Does JS loaded from the server have different privileges to JS entered at the console?

This is great to hear. Growth without a corresponding decline in quality. I must admit, when I read the original article, I was a little skeptical it would work out well.

Are there any design choices you would particularly attribute the success to?

> You'll run into a set of requirements that no one has run into before.

Inevitably. Which means you'll have to resort to "real programming" (as opposed to component assembly) for /that part/ of your project. I don't see this as a fundamental reason why component-assembly can never become viable. As we get better at creating flexible components, these situations will get less common, but they will never go away.

> you'll run into a requirement that's simpler to implement directly than it is to write the glue code for all the components you could use to solve it

100% this. For me this pretty much sums up why component-assembly isn't viable today. For all but the simplest components this turns out to be the case. (e.g. date-picker, file-uploader, or maybe something a bit bigger with /very/ fixed requirements, like a disqus comment trail). But jumping from this to "I don't think it's every going to happen" is overly pessimistic. The glue code is too hard to write? We need a better way to write glue code. That could be a fundamentally different type of language, or a fundamentally different conception of what we mean by "component".

(Aside: In my foolishness I am working on such things).

To date, software engineering has failed to deliver on one critical goal, which is why we are in this mess.

The world of software development needs to be divided in two -- the component creators, and the component assemblers.

To stick with the given analogy, component creators are the people inventing new kinds of plumbing: easier ways of connecting pipes, taps that don't ever drip. Component assemblers are the people fitting out bathrooms.

Creating new components is high-end engineering. It needs to happen far away from the day-to-day challenges of making a client happy.

To a large extent, this division is already present, but it doesn't go far enough. There has been amazing progress - nowadays we work on top of an incredible stack of technology that we don't have to re-invent, but so far we've not achieved the "last mile".

We'll know when we're there because component assembly (i.e. making something for a client) will start to look more like a trade.

Today, just getting a regular been-done-a-million-times-before database-driven website (or whatever), requires FAR too much low-level code. This, I think is what is meant here by "hand made". We should be snapping things together, and often we are, but suddenly you get to an awkward bit and you're back to forging a new kind of pipe joint that never existed before.

For those unfamiliar with ObjectiveC syntax, this[1] is a very nice example, in JavaScript, of how amazing this idea is.

[1] https://github.com/Reactive-Extensions/rxjs-html

I don't see the problem with VAT. Just keep rack of where your customers are and pay the VAT on those in Europe. Where does the payment processor come into it? If you owe VAT then pay it.

Very nice. Another nitpick - I really like what Apple have been trying to do to in creating a UI affordance that says "this is a search box" - i.e. a text entry with round ends. I hope all the devs out there can join in and not use that appearance for regular text fields.