HNHacker News
TopNewBestAskShowJobs

amptorn

700 karma · joined December 6, 2015

submissionscomments
amptorn··on Google Is Battling a Russian Spammer Over the Use of the Letter 'G'
There's a lot of Unicode variations on "google". Potentially too many to make that practical...
amptorn··on Balaji Srinivasan, who may run the FDA for Trump, hates the FDA
Well, when you put it like that, yes, all of them.
amptorn··on Mathematicians bring ocean to life for Disney's 'Moana'
The other thing to remember is that Disney are completely at liberty to reject and re-render scenes when something doesn't look perfect. Arguably, this even applies to examples chosen for the demo video.
amptorn··on Leap second 2017 status
Software still botches leap years, which have been a fact of life for centuries. I think software generally isn't going to get any better.
amptorn··on Macaque monkeys have the anatomy for human speech, so why can’t they speak?
Now I want to know what the album was
amptorn··on C puzzles
When I make a programming language, one of its core design principles will be that there should be no puzzles like this.
amptorn··on Time to Dump Time Zones
...Your solution is to just try at random? And if it doesn't work, give up?
amptorn··on Why GitHub Finally Abandoned Its Bossless Workplace
Well, if their primary source of cash is investors rather than actual business...
amptorn··on 500 Byte Images: The Haiku Vector Icon Format
Get it down to 280 bytes and you can fit one in a tweet: https://github.com/ferno/base65536
amptorn··on Time Zones Aren’t Offsets – Offsets Aren’t Time Zones
> there are historical time zones that aren't representable as whole minute offsets from UTC either.

Luckily we ditched all of those before we invented leap seconds.

amptorn··on Time Zones Aren’t Offsets – Offsets Aren’t Time Zones
What does "am" mean?
amptorn··on Time Zones Aren’t Offsets – Offsets Aren’t Time Zones
Do you know why we invented time zones?
amptorn··on How to destroy the Earth (2006)
This is from 2003, in fact.
amptorn··on Agile Is Dead, Long Live Continuous Delivery
That methodology is awful. And kind of embarrassing.
amptorn··on FBI Paid More Than $1M to Hack San Bernardino iPhone
I still don't believe they've actually hacked it. There's no evidence that they have.
amptorn··on Var and val in Java?
I don't care what you do or how you define them, don't have both `var` and `val` in the same language. Way too easy to confuse.
amptorn··on Require-from-Twitter
One of the interesting side-effects here is that while it encourages you to write very small packages, it also strictly caps the maximum number of packages which you can add as dependencies of your own package. There's only so many `require` calls you can fit in a Tweet.
amptorn··on Stop Using the Daylight Savings Time
I meant the legal definition of time, not the dictionary definition.
amptorn··on Stop Using the Daylight Savings Time
> DST is actually more comfortable to live with by allowing for more daylight after work hours.

Or we could just work slightly different hours? But I guess we're more comfortable altering the definition of time itself...

amptorn··on What the IBM Layoffs Look Like
I'm in the same position, but the presence of all publicity and no real product is part of what leads me to believe that Watson is much like Deep Blue or AlphaGo: it was designed to solve an extremely specific problem, and its capabilities do not generalise readily to any other problem. This despite the fact that Jeopardy is a "general knowledge" quiz.

Imagine if Watson was an app. What would you do with it? As I understand it, it takes an oblique textual reference of some kind and turns it into a concrete, explicit reference. You talk about something and it figures out what you're talking about. That's great! It's a hard problem and it's extremely impressive that it's been solved.

But in practical value terms, that's just a toy. And when would it be more useful than Google Search, or Wikipedia, or Siri? Can Watson do anything, book a trip, set a timer? Well, that capability would have to be plugged in, along with voice recognition and speech synthesis. But can it reliably determine what I want it to do? Or even whether I want it to do anything? Can it follow a conversation? Is it even stateful? What, if any, role does the core Watson functionality play in the systems which eventually do these things?

And how does Watson generalise to other areas? IBM was looking at applying it to healthcare. How would this work? Presumably, here your "oblique textual reference" is a collection of data (symptoms and measurements) collected from a patient, and the end result is a diagnosis, like "What is rubella?". Now, firstly, how useful is that? Is that the hard part of being a doctor? What about carrying out the tests, what about prescribing treatment or choosing further tests? What about missing hints, e.g. can Watson handle visual data from simply looking at a patient? Or does the doctor have to inform Watson that the patient is leaning slightly to her right when she sits, and has trouble getting onto the bed? So does Watson's single trick actually help, does it even slightly alleviate the workload of a medical professional?

And secondly, how readily does Watson adapt from the world of general knowledge quiz answers to the world of medical diagnoses? Are these worlds even remotely structurally similar? Can they be modelled and correlated using the same basic structures? Is there some heavy modification and serious domain knowledge needed to alter Watson to do this? Or are the two problems basically on different planets, such that an entirely different machine needs to be built?

The same goes for applying Watson to business analytics, for example. In my mind, Watson is a machine which, as it currently exists, best case scenario, can look at all of your financial graphs and go "Ding! You're in a recession." And then nothing else. This may be of some use, but I doubt it.

All of this is ignoring the backend, which is that Watson in reality is this monumental pile of expensive, high-performance hardware and IBM Research code - and we all know how much research code ever resembles a working, saleable product. All of that to handle one real-time quiz game. You're going to shrink that until it fits in one office? Or you're going to have a server farm full of them serving requests at great expense?

I feel like people perceive Watson as a hair's breadth from strong AI, and IBM doesn't want to disabuse anybody of that notion. But I think the really valuable stuff in Watson - advances in computing techniques which I don't really know anything about but I'm certain must be there - is far less tangible, and is going to be very difficult to extricate from Watson itself, let alone monetize.

Just off the top of my head. Unsourced hunches.

amptorn··on What the IBM Layoffs Look Like
Watson is a red herring.
amptorn··on Reports Coming in of Big IBM Layoffs Underway in the U.S.
Seems like a stopped clock situation in his case. That prediction is from a year ago and came to relatively little; he didn't predict this at all.
amptorn··on Multiple Black Vehicles with MD License Plates at Apple Executive Briefing Center
Intimidation?
amptorn··on Ask HN: Government law as code?
Yes, you do always need humans to enforce/interpret the laws. Don't be absurd.

The alternative is kind of like that scene in Robocop, except ED-209 is a judge instead of a robot, it sentences you to death rather than killing you directly (the bailiff obeys unquestioningly), and instead of frantically trying to shut the machine down, everybody in the room agrees that you were indeed carrying the gun, and almost certainly still are.

You ask for a retrial and it takes 1/50th of a second to spit out exactly the same response. The rationale for the verdict is a procedurally generated AST running to 5200 pages, which is incoherent not only to you but also to lawyers and programmers alike. (You are given a copy of it on a USB stick, but no computer to read it with.)

A country where laws are applied with absolutely no nuance, context, consideration, empathy or judgement is a petrifying hellscape. A pretty good Black Mirror episode, in fact.

E: this is all ignoring how awful human beings are at software development.

amptorn··on Disabling npm's progress bar yields a 2x npm install speed improvement
"Free software projects with terrible names" is a long, long rabbit hole.
amptorn··on Base65536 – Unicode's answer to Base64
If screen area is your metric, and combining code points are free, I bet you could use [1] to add an arbitrary amount of data to a single character.

[1] http://www.eeemo.net/

amptorn··on Base65536 – Unicode's answer to Base64
Well, null, to start with. Other ASCII control characters. Consecutive line breaks, too.
amptorn··on Base65536 – Unicode's answer to Base64
> Base65536 that isn't actually 65536? Odd.

Well, Base64 isn't actually 64 either.

amptorn··on Base65536 – Unicode's answer to Base64
This is dangerous through, because sending a Unicode string through Twitter subjects it to NFC normalization which can alter its structure. If you want guarantees that this won't happen, your choices of code point are far more constrained, and the unassigned code points, which is most of them, can't even be used. I think this is the purpose of base65536gen.

https://www.npmjs.com/package/base65536gen

← PreviousPage 7 of 7