arscan

David Byrne is currently on tour and my wife took me to the show as a bit of a backup plan (her usual concert-going friends fell through). I think concerts are okay, but I usually get bored after awhile, which is why I serve as a backup guest. I’m much more of a visual person, I suppose.

I was unprepared for how wonderful the show was. I was quite literally tearing up during the very first song from the beauty of the visuals that accompanied the musical performance, and it maintained almost that level for the entire show.

He’s really a generational talent.

Hmmm, I think it’s a bit paradoxical to try to come up with a fun term to encourage the spread of the concept, as the concept itself is quite dull by definition. It’s just software engineering. Boring process stuff where you do a bunch of other things around the ‘building’ part that lets you scale up with quality. Vibe is about fun and go-with-the-flow. This isn’t that. It should connote pocket protectors, not sunglasses. Right?

Clearly I’m not in marketing.

Regardless, I’m delighted that this has gotten people to ‘independently discover’ software engineering best practices on their own.

> but ChatGPT did it in 10min with some back and forth

You might be underestimating the expertise you applied in these 10 minutes. I know I often do.

> it's a toy example

This technology does exceptionally well on toy examples, I think because there are much fewer constraints on acceptable output than ‘real’ examples.

> you don't have to knwo what you're doing to get useful things from AI

You do need to know what is useful though, which can be a surprisingly high bar.

I’m not an expert on American history, but this attitude feels a bit like a modern day adaptation of ‘manifest destiny’.

its pretty easy to dispute and dismiss a single use case for indiscriminate/excessive use of inference to achieve some goal, as you have done here, but its hard to dispute every possible use case

I also think it’s fashionable to have a smaller headcount these days. Historically, the dynamics of businesses encouraged rising headcounts, as ICs weren’t as valued as managers (salary caps basically, as impact for ICs is hard to measure unless you are in sales), and managers generally view headcount as a metric to career and salary growth.

So there was just this general pressure from the middle up to grow instead of paying more to existing staff or finding some other way to spend the money. After all, investors generally want you to spend the money you have access to, otherwise they’ll put it to use elsewhere.

It seems that there is external pressure right now from investors, and on to executives, to push headcounts down as there is a general feeling that good companies should be able to leverage AI to become much more efficient, and higher headcounts just burn money and bog things down. Whether or not that’s true is another question, but the perception exists.

I’m not sure if this is a fundamental change in the dynamic, or just a temporary push against it that will eventually lose steam.

The story that I’ve heard (probably on here) is that the administration did it to make the tax bill look more balanced over the long term by phasing out that tax write-off, while giving them (or the next administration) time to reverse it before it really impacted anything. But, nobody reversed it, until this year.

> Honestly, I was less than half as useful as the current crop of robots and they still tolerated (in fact, encouraged) my involvement.

Don’t sell yourself short. External contributors are extremely valuable as they are often users and provide real world validation of a need for whatever they are contributing. They also retain any knowledge for any feedback they receive that they can apply to future contributions. And they also become advocates for that software, helping it grow its user base.

LLMs, not so much.

Thanks for this. Browsing the latest sierra, microprose, etc games at the local egghead store was my absolute favorite thing to do as a kid. It was in a strip mall next to an actual mall, so my parents would get me to go along on shopping trips as long as we could swing by the egghead to see what was new on the way home.

The feedback loop from potential developer users of your API is excruciatingly slow and typically not a process that an API developer would want to engage in. Recruit a bunch of developers to read the docs and try it out? See how they used it after days/weeks? Ask them what they had trouble with? Organize a hackathon? Yuck. AI, on the other hand, gives you immediate feedback as to the usability of your “UAI”. It makes something, in under a minute, and you can see what mistakes it made. After you make improvements to the docs or API itself, you can effectively wipe its memory by cleaning out the context, and see if what you did helped. It’s the difference between debugging a punchcard based computing system and one that has a fully featured repl.

> Protocols and standards like HTML built around "be liberal with what you accept" have turned out to be a real nightmare.

This feels a bit like the setup to the “But you have heard of me” joke in Pirates of the Caribbean [2003].

I have many fond memories of this game. Of course it had all the issues that existed in that genre — in particular it was easy to get stuck… forever. But being a kid back then with a lot of free time and not many other distractions it was great.

A danger with the arrangement of this article is that it takes awhile to get to the point, which actually in line with your view. He hints at it in the title and the very next paragraph, but maybe you didn’t get that far?

> And this week, I want to share with you how I did it, what I learned, and why I think self-hosting is NOT the future we should be fighting for.

I’ve been thinking of LLMs a bit like a credit-card-for-the-mind, it reduces friction to accessing and enabling your own expertise. But if you don’t have that expertise already, be careful, eventually it’ll catch up to you and a big bill will be due.

Say what you will about the language, but I’ve come to appreciate how important it is to stick up for your users/customers, which is what Linus is doing here. It’s not ok to inconvenience them without an extraordinarily good reason to do so, and it absolutely is not ok to blame them.

I’m sure my reaction to this was different when I first read it many years ago. But now, I can strongly relate to the feeling behind these words.

An aside, but it’s quite unfortunate that logger.info and logging.info are automatically linkified because of the .info TLD in this case. I don’t recommend clicking on those links.

> Looking at the kind of mistakes Claude made, it seemed like a human could make those mistakes too (very useful feedback for us to improve our API ).

This is something we've found for our API -- just having LLMs attempt to use it helps us identify things that we haven't documented well or placed enough emphasis on (for things that are critical but are non-obvious or may be drowned out by other less important information). Improvements that help the LLM tend to be good for developers too.

This setup is also compatible with GitHub CodeSpaces, which I think is an exceptionally nice development experience if you are on something like a Chromebook, and gives you isolation from your local network. Also lets you share your app running in dev-mode with anyone for feedback, without having to poke holes in your network or whatever. Also also, you can use the Cline extension in this environment if you still want the cursor-style copilotey experience.

There was an episode of Car Talk where a caller described this exact situation. In that case, for the model in question, there were only so many varieties of keys made, so you could use your key on some other cars.

There was a time when Microsoft Word would treat the passive voice in your writing with the same level of severity as spelling errors or major grammatical mistakes. Drove me absolutely nuts in high school.

What is the temperature on either side of this “wall”? My mental model here, which is probably incorrect, is that the “temperature” on the outside of the wall could be higher but the density is much lower, thus even less heat transfer going on (but, still, high energy particles that can hit you, registering a high temperature). I get all kinds of mixed up regarding the difference between heat transfer and measured temperature.

In other words, people are discovering the value of standard software engineering practices. Which, I think is a good thing.

FWIW, the 19th century style dataset this is trained on doesn't seem to have any examples of delve [1], with the exception of:

> It was into that bank that the creature had worked its way, and on listening I could hear it delving and scraping at a great rate, about a yard from the back of the wall.

I bring that up to point out that this isn't necessarily (more) common in the 19th century style print literature, so the observation might not be silly. The model creating the modern synthetic version injected 'delve' 9 times, which implies that it is more frequently used in modern literature or just something that models tend to inject. Though, I could be missing something (either in searching the data set, or how this works).

[1] https://huggingface.co/datasets/dleemiller/irish_penny_journ...

I very much agree with this. Find a job where you can meaningfully contribute to a product that is important to your company (generates a meaningful amount of revenue), and after a little while the natural thing that happens is you become irreplaceable because of that retained value in your head.

It doesn’t have to happen, and with some effort can be somewhat avoided, but it’s the default outcome. Depending on your goals and career aspirations, this can be a wonderful thing, or it can be a bit of a curse.

And for the curious, this current iteration of AI is an amazing teacher, and makes a world-class education much more accessible. I think (hope) this will offset any kind of over-intellectual dependence that others form on this technology.

Very cool, the interactivity of this makes this a much better learning tool than a set of static images (for me, at least).

One minor suggestion: on mobile put the date scrubber on the bottom, otherwise my thumb gets in the way of the UI while sliding back and forth through time :)

Also, I’m not sure if a log scale for time makes sense in this case. It confused me for a second, at least.

Great job, thanks for sharing!

Hah, this is about the ~2010 Google ai contest “planet wars”. I participated in the 2011 iteration, ‘ants’, which was a fun but humbling experience. Anyone else try that? I recall the final winner’s solution was far less elegant than I expected after the code was released, but man was it effective.

Good ones are expensive to create and turns out there isn’t that much money in it. It wasn’t clear this was the case early on when HTML5 came out and really enabled these experiences. But after you make a few and realize how much goes into creating them, and how hard it is to extract value from them, it doesn’t make that much sense.

Also, that us election needle from 2016 really turned a lot of people off of to the whole genre, I think.

To technical people, sure. I don’t think the average person knows about provenance rules of subdomains though and how it’s useful… it’s more just a bunch of symbols they don’t care about.

And we understand the threats here… a very real problem is someone forgetting to renew one of these .org or .com domains (maybe the person that maintains it retired) and a malicious actor grabs it after expiration, stands up a scraped copy, and uses it to collect parking ticket payments or whatever.

I was actually thinking a bit more about the diversity of domain names under .gov, though I realize now that the parent comment I replied to was about .org and .coms. I think you get a bit of those provenance assurances if they are under .gov, as a practical matter it’s harder for malicious actors to own one of those than one under other tlds. And then instead of forcing a strict taxonomy that is mostly for the benefit of the infrastructure maintainers (very enterprise software), there is freedom to use a name that makes the most sense for the target user.

URLs are a part of the UX of websites. The domain often represents the first interaction between the user and the site. Domains that follow a strict hierarchical structure that aligns to some real-world hierarchy may not be the best first interaction with the user, or at least not in the opinion of those that are creating the site.

So, I think it’s natural for site owners to want this freedom. Then it comes down to whether there should be constraints forced on them or not by policy for some greater good. In the US, generally, central planning on this type of stuff isn’t really part of the culture.