jkaptur

(I'm not an expert. Take all of this with a grain of salt and just imagine I'm sticking "I think" in front of every sentence.)

It's easy (O(path length)) to check if a (target, path, tuple) is valid, but the paths themselves are of phenomenal lengths:

"He developed a procedure for constructing systems in which the fastest way to determine whether one state is reachable from another is to map out a sequence of transitions between them. That allowed him to use the length of the shortest path between two carefully chosen states as a measure of the difficulty of the reachability problem."

That's the guarantee: we can prove you have to do a ton of computation because you have to traverse this huge path. That's why this whole area of study isn't tainted by the question of whether P=NP. (Formally, because it doesn't have a polynomial-size witness).

So I think you could make a cryptosystem out of this, but, since you'd (for practical reasons) use relatively short paths, you wouldn't get the difficulty guarantees of this research.

I absolutely agree, but I find the situation incredibly funny. There are three characters here: me, the LLM, and the compiler. Two of them are robots, but they refuse to talk to each other - it's up to me to tell the LLM the bad news from the compiler.

Right! Always Sunny is funny (the bells in that scene get me every time), but, in the name of comedy, it skips over the obvious: you can go to natural history museums! They frequently have talks where you can meet people who have dug dinosaur bones out of the ground, and they love to answer all your questions about it.

Sure, they could all be faking it, but at that point you have to question how much evidence it takes to believe in anything at all.

They're really no better than plenty of reasonably-sized cars.

Clippy was hated whether he was right or not. Consider the most popular catchphrase "It looks like you're writing a letter..."

In my experience, Clippy was never wrong about that - precision and recall were absolutely perfect - but people do not appreciate being interrupted, particularly at the precise moment their work has begun.

The hard part of creating an assistant is not coming up with some stuff to suggest, it's coming up with a time, place, and manner to make those suggestions so that users might be open to them.

> creating an assistant out of it will someday become as easy as keeping a Juypter notebook running.

Absolutely true, but if you intend for that assistant to proactively inject itself into people's lives, get ready to be held to an unimaginably high bar of quality.

At the time it was recorded, that was cutting edge tech, wasn't it?

This is not a huge surprise. They've been extremely behind schedule putting out print issues for quite some time. I hope they can turn things around soon - I'm a big fan.

To be fair, that's basically the only way to play Diplomacy.

https://grantland.com/features/diplomacy-the-board-game-of-t...

Here's a skull now; this skull has lain in the earth three and twenty years.

For those just looking to dip a toe in different layouts: I remapped movement keys and it's awesome. I made a demo so you can see for yourself: jkaptur.com/layout/

So the attack goes: 1) compromise some site to serve arbitrary JS 2) have it serve simple JS that requests other JS that contains the real malicious payload.

And the reason for this two-step architecture is to make it convenient to change the real payload.

And the problem is where to host the real payload. The first idea was Cloudflare, but Cloudflare keeps taking that sort of thing down. So now they host it "on the blockchain" which means it "can’t be blocked".

What I don't understand is who is actually serving the HTTP call that effectively proxies the data from the blockchain. It seems like they (like Cloudflare, or a regular hosting service) are opening themselves up to all sorts of risks by serving arbitrary content.

Hyper-pedantic note: I think "sigmoid" is the term that describes these curves in the full generality. https://en.wikipedia.org/wiki/Sigmoid_function

Ah, thanks for explaining.

Thanks for clarifying!

"Microsoft.... built a custom XML tool into its word processor in 2007... this was a tool for power users, and was only used by a small percentage of its user base."

I'm definitely confused by that statement and its link, because it implies the relevant tool is the disk format for every Office file, which has been described by an Excel program manager as "complicated enough to reduce a grown programmer to tears." https://www.joelonsoftware.com/2008/02/19/why-are-the-micros...

I had a job building cool looking dashboards, and one day the users asked me to add a column where they could put "functions, like SUM and so on".

I quit shortly thereafter and got a job on the Google Sheets team.

So much effort for a joke - you really have to take your hat off to the team.

My local library is already exactly this. It's great, but they get no credit for it.

We'll just call it "NYC" ;)

It's definitely interesting how people with the same title can have such different experiences. I'll try to give the other perspective, but please understand that it's a huge topic and this is a simplification.

My experience has been an issue is raised - usually by management or "the business", but sometimes by engineering. "Our website is too slow!". Now someone has to step up and clarify things. Like: * what does "fast" even mean for our website? It would be great to have an objective metric. * Is there such a thing as "fast enough"? * How much time / effort / money is a given improvement worth? * How do we prioritize this vs. everything else we could be working on?

A project spec comes out of those discussions, but the discussions themselves have significant technical and business inputs.

I don't view taking part in these discussions as "extra" responsibility, I view it as a core part of the job. I've also seen what happens when specs are written without technical input: it makes implementing them a lot less pleasant.

I tried this. A very large company reached out and asked for a custom version of diffdiff.net. They asked what my hourly rate was, and (thinking along the lines of this article) I offered to do it for a flat rate instead.

They ghosted me!

In retrospect, a "very large company" didn't reach out, a relatively junior person at the company did. I should have worked within their constraints. (Or just not taken the job, which, I guess, is effectively what I did).

I served on a jury and you described it perfectly: imperfect people took the job seriously, interpreted our understanding of the law and the facts of the case, and listened to each other, both lawyers, and the judge.

I've often been tempted by the "delete your flaky tests" extremist sect of that religion :)

> for no discernible reason at all

Are you asking? Post the location and I'd be happy to weigh in from a cyclist's point of view.

Features interact combinatorially, which introduces complexity.

Faxing isn't a great example, since from the perspective of most other features, it's probably just a special case of printing, but you can imagine it made it harder to build a web-based version of Word (what number are you faxing from? How does the computer in the MS data center contact or emulate it?)

One line that demonstrates the agony and the ecstasy of simplicity: $ man ls | grep "\-\w," | wc -l

On the one hand, beautiful, recursive composability of small, universal tools.

On the other hand, the tool we're introspecting is 'ls'. If we truly valued simplicity, ls would have no flags - you would pass its output to awk to extract the data you want.

A more thoughtful writeup: https://danluu.com/cli-complexity/

Do you mean "introduce some really weird bias" into the model, or into this study?

Yep. Also learn like 5 data structures and 7 algorithms in case the relevant manager can't just unilaterally hire you.

The price of Campbell's soup began its dramatic rise after Warhol's record-breaking sale of Campbell's Soup Cans in 1970. (https://en.wikipedia.org/wiki/Campbell%27s_Soup_Cans)

Although back then you could buy Campbell's Soup Cans (the painting) for just $60,000. In 2022, a famous Warhol painting went for ~$195,000,000.

Well, it describes what it's like to be in system B. All it says about how to get in is to build something (good? great? flashy?) and hope someone notices.

This clearly does work sometimes, but it's not necessarily the best advice for everyone.