Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
pmarreck
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
42 ms
·
841.
▲
by
pmarreck
2y ago
Sure. We have a situation where 1) the "West" is constantly at war with Islamists and their discontents in fights almost exclusively begun by them (I believe they are religion-motivated, based on my research, but that's ano
842.
▲
by
pmarreck
2y ago
See my edited response (that you replied to)
843.
▲
by
pmarreck
2y ago
I have a toddler and zero assistance other than daycare. IYKYK.
844.
▲
by
pmarreck
2y ago
Yes. I'm well-aware of the analogy of the slowly boiled frog, etc. But it's also the slippery-slope fallacy, so... Why are we assuming facts here before we know of them? She is a Turkish national on a student visa. Those folks do
845.
▲
by
pmarreck
2y ago
FYI roc is in the early stages of a complete rewrite from Rust to Zig and... that may take a while. I'm messing with Nim in the meantime for some commandline tooling
846.
▲
by
pmarreck
2y ago
You got me thinking again about this (even days later) so I asked ChatGPT to get some insight and man, not only did it make your point clearer, it made MY point clearer TO ME. In lieu of me. While also teasing me lightly about it. LOL, that
847.
▲
by
pmarreck
2y ago
Great points all, admittedly! I think I'm scarred by having churned too many man-hours supporting spagetti ASP code and Ruby "magic" behavior.
848.
▲
by
pmarreck
2y ago
Why? If it's already out there, it's kind of too late, is it not?
849.
▲
by
pmarreck
2y ago
I love that ABC thing and will be forced to remember it! Good counterpoints. I did investigate it just a few months ago but then found criticisms about its translation of V to C being quite subpar, but eh
850.
▲
by
pmarreck
2y ago
> through the overuse of ‘mate’ The whole thing seems Oz-influenced (example, "in the bush with the roos"), which implies to me that he's prompted it to speak that way. So, you assumed an error when it probably wasn't
851.
▲
by
pmarreck
2y ago
fair points. looking back on this comment from 3 days ago... all I can say is that my toddler kid isn't sleeping through any night consistently :/
852.
▲
by
pmarreck
2y ago
Fair enough, but I also know a few guys who started in PHP and then never left it, and I kind of feel bad for them
853.
▲
by
pmarreck
2y ago
> There are far worse jobs out there. Fair enough. But there are also far better. I can't help caring about this one. It's a curse. lol
854.
▲
by
pmarreck
2y ago
> It is completely possible to have a PHP application written 3+ years ago that is well maintained, continues to promote best practices, has excellent test coverage, that isn’t a nightmare to deal with Of course, as soon as you depend on
855.
▲
by
pmarreck
2y ago
> literally every language could have a book written about its problems OK, Elixir's been out for quite a few years now and is used by thousands of people at this point. Find me the "elixir sucks" webpage, which is super e
856.
▲
by
pmarreck
2y ago
Good point on the type thing. Not so much on the "everything has a few warts after 30 years". Many of the warts were baked-in at the start or in the very early days (like in many other languages), which is why language design is s
857.
▲
by
pmarreck
2y ago
https://chatgpt.com/share/67e29912-bc64-8004-be75-3d0df11d07... I have a toddler demanding 99% of my free time so I apologize in advance for the LLM shortcut but this looks correct based on my experience
858.
▲
by
pmarreck
2y ago
https://chatgpt.com/share/67e29912-bc64-8004-be75-3d0df11d07... I have a toddler demanding 99% of my free time so I apologize in advance for the LLM shortcut but this looks correct based on my experience
859.
▲
by
pmarreck
2y ago
keep looking and you’ll see why. chaotic-neutral dictator/chief dev categorizes any criticism (regardless of merit) as trolling and bans; language design far exceeds technical merit of implementation details; basically, things only loo
860.
▲
by
pmarreck
2y ago
Summary for the time-pressed: Large Reasoning Models (LRMs) are a new type of AI system designed to perform specific types of reasoning. They generate “chains of thought” to solve problems, mimicking human reasoning processes. While LRMs ha
861.
▲
by
pmarreck
2y ago
If it's about collections of things, use collection iterators (`each`, etc.) and avoid the problem entirely.
862.
▲
by
pmarreck
2y ago
> I would not describe current LLMs as able to generate essays on anything Are you invisibly qualifying this as the inability to generate interesting or entertaining essays? Because it will certainly output mostly-factual, vanilla on
863.
▲
by
pmarreck
2y ago
https://transformer-circuits.pub/2024/scaling-monosemanticit... that is the paper being linked to by the "marketing material". Right at the top, in plain sight. If you were arguing in good faith, you'd h
864.
▲
by
pmarreck
2y ago
It's more or less always been like this. It's why I moved to mainly backend (where feasible). Because someone has to maintain that crap at some point.
865.
▲
by
pmarreck
2y ago
Yep, that's what I expected, thanks for filling me in.
866.
▲
by
pmarreck
2y ago
> tend to do worse That's a subjective evaluation, no? ;) I'm sure their parents would say they did way better on the religion/dogmatism ruler! > So many families around us homeschool This seems like the "gateway d
867.
▲
by
pmarreck
2y ago
100% correct! Biggest mistake ever was not to grow it organically/incrementally while maintaining backwards compatibility!
868.
▲
by
pmarreck
2y ago
Deterministic random value dithering, where the chance of being the dithered color or not is based on the percentage that the true value is that color?
869.
▲
by
pmarreck
2y ago
If only IPv6 had taken off...
870.
▲
by
pmarreck
2y ago
would a widescreen make this better or worse? (I like to work in UWQHD.)
More ›