Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
josefrichter
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
6 ms
·
61.
▲
by
josefrichter
7mo ago
scared to go check my washing machine display
62.
▲
by
josefrichter
7mo ago
this is wild.
63.
▲
by
josefrichter
7mo ago
I wrote my university thesis 25 years ago on automating some parts of art traffic management. The fact that it's "humans over radio" coordinating everything sounded ridiculous already back then. but it's still "don&
64.
▲
by
josefrichter
8mo ago
Is it really officially called "Department of War" now?
65.
▲
by
josefrichter
8mo ago
Use Elixir. Not exotic at all.
66.
▲
by
josefrichter
8mo ago
We can argue about percentage points and months here and there, but we're very quickly coming to 100% and people aren't even afraid to admit it publicly anymore. The prophecy was definitely on point.
67.
▲
by
josefrichter
8mo ago
It feels like it's Saturday and HN is full of scared blog posts.
68.
▲
by
josefrichter
8mo ago
This mentality never worked in IT world. We've always had high pace of change and endless learning and adaptation to new tools and approaches.
69.
▲
by
josefrichter
8mo ago
"Anthropic’s CEO predicted AI would write 90% of code within three to six months of March 2025. None of this happened as predicted." – oh it absolutely did happen!
70.
▲
by
josefrichter
8mo ago
Radically different? Sounds to me like the standard spec driven approach that plenty of people use. I prefer iterative approach. LLMs give you incredible speed to try different approaches and inform your decisions. I don’t think you can eve
71.
▲
by
josefrichter
8mo ago
These are very broad generalizations and accusations based on very few individual cases, each of which has its own specific context. And "expressing points of view alternative to position of Brussels and Berlin" sounds like typica
72.
▲
by
josefrichter
8mo ago
Not true at all with frontier models in last ~6 months or so. The frontier models today produce code better than 90% of junior to mid-level human developers.
73.
▲
by
josefrichter
8mo ago
Exactly
74.
▲
by
josefrichter
8mo ago
AI most definitely is a coworker already. You do delegate some work for which you previously had to hire humans.
75.
▲
by
josefrichter
8mo ago
I'm in Czechia, next to Germany. Just opened Ria Novosti and Russia Today in two other tabs, nothing blocked here.
76.
▲
by
josefrichter
8mo ago
For Europeans? They don’t need anything like this, zero benefit. May benefit someone in North Korea, China or the United States.
77.
▲
by
josefrichter
8mo ago
I’m in the EU and haven’t encountered any of these, except the copyright restrictions - which is really a different matter.
78.
▲
by
josefrichter
8mo ago
Very useful for supervising Claude and occasionally steering it. Now that Claude finally auto-suggests next command, it partly offsets one of the biggest obstacles of mobile phones - typing on tiny keyboard is painful.
79.
▲
by
josefrichter
8mo ago
A bit younger, and exact opposite. Probably the most excited I've ever been about the state of development!
80.
▲
by
josefrichter
8mo ago
It sounds to me like an exercise in wordplay within very broad definitions. My initial reaction was towards the claim "It's literally the same thing as async await in python and nodejs" which is only true in such case.
81.
▲
by
josefrichter
8mo ago
No, it’s not literally the same thing. I’m afraid you greatly underestimate and fundamentally misunderstand Erlang processes by making such comparison.
82.
▲
by
josefrichter
8mo ago
Exactly. Basically their thought are invalidated often quicker than they hit the publish button.
83.
▲
by
josefrichter
8mo ago
Come on, this shows fundamental lack of understanding and experience on your side.
84.
▲
by
josefrichter
8mo ago
The OP’s example of AI writing 500 LOC, then deleting 400, and saying it didn’t… Last time I saw something like that was at least a year ago, or maybe form some weaker models. It seems to me the problem with articles like this is while they
85.
▲
by
josefrichter
9mo ago
I am not into conspiracy theories, but I find it very unlikely that our governments can’t read all our messages across platforms.
86.
▲
by
josefrichter
9mo ago
You can organize things. It's surprisingly easy. You just put up a FB event. When I was younger and moved to a new (foreign) city, The first thing I did was to create a "picnic" for people coming from my country. No agenda, n
87.
▲
by
josefrichter
9mo ago
Isn't the suggested 'path forward' basically a description of Elixir/Erlang BEAM?
88.
▲
by
josefrichter
9mo ago
The murder of Twitter seems to be a part of a greater scheme of things.
89.
▲
by
josefrichter
9mo ago
The "AI stealing jobs" is just a new-age Luddism. Yes, of course it will be painful for many, but at larger scale, even those who lose the jobs should benefit overall. "your ability to create a mental representation of the pr
90.
▲
by
josefrichter
9mo ago
Didn't he basically observe the second law of thermodynamics? The Second Law says an imbalance (like heat in one corner of a room) will "solve itself" by spreading out until the room is balanced. The Napoleon Technique assume
More ›