HNHacker News
TopNewBestAskShowJobs

jw1224

1,959 karma · joined March 20, 2017

submissionscomments
jw1224··on GPT-4 is not getting worse
Are you using the API, or just the ChatGPT UI? Do custom instructions/system messages help at all?

Saving resources like this seems okay(ish) for general-purpose ChatGPT, as they'd know there's a human who can interpret the abridged output. But for API-driven requests this would be pretty terrible, and makes it far harder to develop with.

jw1224··on GPT-4 is not getting worse
This was using gpt-4-0314, fixed and datestamped. The counting shortcut still affects newer models too, but the old version undoubtedly changed behaviour without warning.
jw1224··on GPT-4 is not getting worse
I first noticed the counting shortcut back in April. It still persists with both the previous and current models.
jw1224··on GPT-4 is not getting worse
Yes, it seems their servers are terminating responses early. Unfortunately it's undocumented, was introduced unexpectedly one day, and prevents full use of larger context-window models.

It also charges in full for the entire response, which seems to continue running in the background. Definitely a bug at their end.

jw1224··on GPT-4 is not getting worse
There is an ongoing bug[1] with OpenAI's API, where it stops streaming responses after exactly 5 minutes. When I first came across the issue, I debugged it by writing a prompt along these lines:

  > Output every number from 1 to 10,000 in written form (e.g. "one", "two", etc.). Respond with one number per line in numeric order.
As expected, the API would begin counting every number just as I asked. This would continue until exactly 5 minutes, when the stream would abruptly halt. Using this technique I was able to identify the bug. Every few weeks I run this test again to see if it's fixed (it broke something in production for me), but the bug remains open.

However, after a couple of months, the exact same test became useless. The model began taking "shortcuts", and would respond along these lines:

  > four hundred and twenty eight
  > four hundred and twenty nine
  > [...]
  > nine thousand nine hundred and ninety eight
  > nine thousand nine hundred and ninety nine
  > ten thousand
Yep, it literally started just writing "[...]" and skipped all the way to the end.

When instructional determinism is reduced like this, it's impossible to say it's not getting worse :(

[1] https://github.com/openai/openai-python/issues/399

jw1224··on An Internet of PHP
> But the success of PHP comes from the people who like PHP the way it was

I'm sure that was the case for many years, but I believe the successful resurgence of modern PHP is coming from the people who want to see PHP actively improving. Not those who like PHP the way it was!

PHP got a nasty reputation thanks to things like the "Fractal of Bad Design" site. Modern PHP has resolved (effectively) all of those early critiques.

The ongoing success of PHP comes from it continually improving :)

jw1224··on An Internet of PHP
Laravel has a dedicated marketing page especially about this:

https://laravel.com/frontend

I personally use different approaches in different projects, but it works great both ways. Inertia is really excellent with frontend frameworks like React/Svelte, whilst Blade is brilliant for server-side templating too.

jw1224··on An Internet of PHP
> Doing one simple thing seems to take a mountain of incomprehensible overhead.

If you just need to do one simple thing, don’t use a framework :)

If you need to lots of simple (or complex) things, in an organised manner across a single project, Laravel is great!

(Having said that if you have an example of what you were referring to, I’d be curious)

jw1224··on An Internet of PHP
It’s the name of a PhpStorm plugin, which adds extended support for Laravel. PhpStorm is built by JetBrains on IDEA.
jw1224··on An Internet of PHP
PHP dev since 5.0 here. I think the severity of these breaking changes is overstated.

I’ve been able to keep several large business-critical projects up-to-date for years, even up to PHP 8.3, with little to no breaking changes encountered.

PHP 8 didn’t really break adding properties to objects dynamically, it just deprecated one bad-practice way of doing so. I remember worrying about this before upgrading, across my entire project with 40+ Composer direct dependencies, there were zero instances of this deprecated approach being used.

With other breaking changes there’s plenty of notice given (years, even). You can control where and how deprecation notices are logged, there really shouldn’t be any surprises at upgrade time.

“Old PHP” made it easy to write buggy or unpredictable code. Modern PHP feels more like writing TypeScript vs JS, there’s real safety and a much better developer experience now. PHP is having a resurgence for good reason, yes, it’s (marginally) harder to keep up with, but with proper tooling (PhpStorm, phpstan, etc.) the DX is leaps and bounds ahead of where it used to be.

I thoroughly enjoy writing modern PHP. I don’t have to worry about the language doing unpredictable dynamic type casting unless I want it to explicitly. The flexibility of old PHP is still there, but it’s much harder to shoot yourself in the foot accidentally.

jw1224··on Procreate Dreams
It’s sad to see blanket statements like this. Stable Diffusion can be an extraordinarily powerful tool for creatives.

I know two professional artists who both hated the concept of AI image generators. But since Photoshop added Generative Fill, I know both have enjoyed using it, and I’ve not heard another complaint since.

I’d bet Procreate add AI capabilities sooner or later. Once you’ve tried it in Photoshop, it’s impossible to look back.

jw1224··on Special Relativity: Lorentz Transformations (2018) [video]
ScienceClic is an incredible and underrated channel, glad to see it shared here!
jw1224··on Uninstall the NightOwl app
> Huh? Setting a schedule/location for nightshift and setting the dark mode setting to auto will always change instantly

Not in my case?! I’d say there’s a 25% chance that Dark Mode enables at sunset. It’s been this way for years — even up til Ventura.

Has it worked flawlessly for everyone else the whole time?

jw1224··on Show HN: Axilla – Open-source TypeScript framework for LLM apps
Two great Phish songs!
jw1224··on “The Onion” founder explains his strategy for sparking creativity
I’m guessing you’re referring to their YouTube channel, in which case, I totally agree. But AFAIK the website is still just as good as always.

They used to put out brilliant high-production value videos like this one: https://youtu.be/Y2j3PsDW33E

Now it’s nearly all low-effort YT Shorts like these… https://youtu.be/-0CVJS1FTmY

Perhaps the satire has just gone over my head? But it’s hard to give them the benefit of the doubt when all their videos are like this now…

jw1224··on Huazhong University demonstrates LK-99 diamagnetism
That's what makes prediction markets work — if you (or anyone else, for that matter) feel the market has priced the expected outcome wrong, you stand to make a fortune!

Great article on the economic theories behind it here: https://astralcodexten.substack.com/p/prediction-market-faq

jw1224··on Marijuana addiction: those struggling often face skepticism
Perhaps (respectfully) this is not good advice? In my experience a “sense of control” over one’s addiction only serves to keep it alive.
jw1224··on NativePHP: A framework for building desktop apps using PHP
Laravel has a dedicated marketing page specifically targeted towards JS-first developers:

https://laravel.com/frontend

jw1224··on Sweden Sans
> CONTRIBUTING.md
jw1224··on Show HN: Peeng – like Pingdom, but the other way around and simpler
Whilst it can be used that way, peng more broadly means something is generally attractive / appealing / impressive. It doesn’t necessarily have sexual connotations.
jw1224··on Pines – Alpine and Tailwind UI Library
Each component is so simple, there’s no upgrading to be done. It’s just an handful of Alpine directives, most don’t include more than a line or two of JavaScript.

Just copy-paste into your project, tweak as needed, and that’s it.

jw1224··on Pines – Alpine and Tailwind UI Library
It’s great for quickly adding interactivity to any static HTML element.

You can turn any element into an “Alpine component” simply by adding the x-data attribute.

Imagine a drop-down menu like so:

  <div>
    <button>Toggle menu</button>
    <div>
      Menu item 1…
      Menu item 2…
      etc…
    </div>
  </div>
Without JS, everything would be visible at all times.

You’d make this reactive with Alpine like this:

  <div x-data=“menuVisible = false”>
    <button @click=“menuVisible = !menuVisible”>Toggle menu</button>
    <div x-cloak x-show=“menuVisible” x-transition @click.outside=“menuVisible = false”>
      Menu item 1…
      Menu item 2…
      etc…
    </div>
  </div>
When the button is clicked, the menu opens (or closes). x-transition adds a smooth fade effect, @click.outside hides the menu when clicked away from the element, and x-cloak prevents it from appearing before the DOM has fully loaded.

Those are just a few of the most common Alpine directives. It doesn’t try to do everything, just the basics simply and efficiently.

I use it for most projects now.

jw1224··on So long “prompt engineering,” we hardly knew ya
This is just how language evolves, stop gatekeeping it. Most engineers these days have never been near an actual engine, and nor do they have any need to.
jw1224··on So long “prompt engineering,” we hardly knew ya
The culture war surrounding prompt engineering is so dumb. My eyes roll hard into the back of my head every time I read a comment like this.

Prompt engineering is not the same as programming. In some regards, it is “better”. In other regards, it is “worse”. They’re two different yet similar disciplines, each with their own strengths and weaknesses, but both equally legitimate.

Programming is simply the skill of being able to communicate with a compiler/interpreter effectively.

Prompt engineering is simply the skill of being able to communicate with an LLM effectively.

I wish we weren’t so divided over an incredible new technology.

jw1224··on Real-time messaging
This seems complex, and not necessarily in a good way
jw1224··on “WebAssembly runtimes will replace container-based runtimes by 2030”
WASM is a compiler target for countless languages, unlike JVM which is tied to Java. That’s what makes it incomparable to JVM… It’s a global runtime-on-demand for whatever you throw at it.
jw1224··on “WebAssembly runtimes will replace container-based runtimes by 2030”
It makes sense. Why worry about hosting your own server, when you can just execute one locally in the visitor’s browser?
jw1224··on Comparing Adobe Firefly, Dalle-2, and OpenJourney
That’s not a fair comparison, as Midjourney is outstanding at a wide range of styles beyond photography.

Generating a “word bubble” is going to look terrible in every major diffusion model. Cohesive words and writing in image models is still highly specialised.

jw1224··on GPT-4 apparently can't recite the litany against fear from Dune
After working professionally for several months with the versioned models on a daily basis, I can say with absolutely certainty they’re being changed behind-the-scenes.
jw1224··on GPT-4 apparently can't recite the litany against fear from Dune
The API suggests this is intentionally, possible due to a (new?) copyright filter:

https://twitter.com/arkephi/status/1670176857394741254

← PreviousPage 3 of 11Next →