Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
bad_username
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
7 ms
·
91.
▲
by
bad_username
5mo ago
The article applies to a narrow case of a totally green field application that's going to be completely vibecoded. This is the only case where you reasonably can be indifferent to what the language is, and so you can abandon familiar P
92.
▲
by
bad_username
5mo ago
This is correct. Substack was one of the first "mainstreamish" platforms to go "against the grain" and try to be neutral with respect to the topics and ideas it platformed. Notably, it allowed deviations from the "p
93.
▲
by
bad_username
5mo ago
True. But this problem is fixable individually with a smog mask outdoors and air purifier indooes. Systemic issues in other countries are rarely so easy to work around.
94.
▲
by
bad_username
5mo ago
Film grain is great for dealing with crappy noise reduction, but I found no good fix for oversharpening yet. Gaussian blur doesn't do it.
95.
▲
by
bad_username
5mo ago
I never particularly revered Dawkins, but he always ranked reasonably high in educated circles, and his books were very popular. But my goodness, the buckets of crap that get poured on him nowadays, in stark contrast with even 10 years ago.
96.
▲
by
bad_username
6mo ago
I am in Europe and The Guardian is quite visibly left-leaning.
97.
▲
by
bad_username
6mo ago
The efforts to divide and weaken the western world have been so massively effective that I sincerely wonder if there is not a state-sized adversary orchestrating all of it for years or even decades. The myopic glee with which people deconst
98.
▲
by
bad_username
6mo ago
I have tons of apps I installed (mostly from Play Store) since like 2012, and that were grandfathered in through Samsung Switch from phone to phone as I replaced them with one another. A lot of data in them, too. Will they, and the data, ju
99.
▲
by
bad_username
6mo ago
John Cleese suggested something similar when solving hard problems that require creativity. https://youtu.be/Pb5oIIPO62g?si=qML6bM5brI_XES9l
100.
▲
by
bad_username
6mo ago
> Generative AI tools, .. supercharge the spread of disinformation and fascism, ... and concentrate wealth in fewer hands People caught up in this line of beliefs generally tend to be more neurotic and unhappy about most things.
101.
▲
by
bad_username
6mo ago
Speak at the same time, but in a slightly different place? If you really care about your speech (and not silencing the speech of others), that is clearly acceptable.
102.
▲
by
bad_username
6mo ago
Where violence is acceptable as a tool, it empowers "cruel humans" on average much more than "beaten dogs".
103.
▲
by
bad_username
6mo ago
> Large language models can’t do much with raw notes or scattered documents. Actually, they thrive with raw notes and scattered documents. I hoard project information in Obsidian with deliberately very little structure: everything lives
104.
▲
by
bad_username
6mo ago
> It had a multiracial cast and female officers. True. And yet it was not woke.
105.
▲
by
bad_username
6mo ago
No, the outright political warfare in the US you are alluding to began much earlier than 2020. Things were getting quite ulhealthy already back in 2015/2016, probably even earlier. The "black swan" for this deep division was
106.
▲
by
bad_username
6mo ago
> It’s just incapable of the thing that makes a real architect valuable: saying “no.” I have had Claude many times tell me "not really, here's a better way" when asked "hey Claude, is this a good idea?". Granted,
107.
▲
by
bad_username
6mo ago
This is why legacy media keep losing audience and trust.
108.
▲
by
bad_username
6mo ago
This is a reason why, before LLMs truly become contenders for replacing humans, the sharp distinction between pre-training, fine-tuning, and providing context in a conversation has to disappear. The model must be able to update itself (lear
109.
▲
by
bad_username
7mo ago
I do not think most executives are particularly enamored with AI. They are being mostly driven by the fear of missing out. More precisely, their thought process is: if they bet on AI and fail, they can plausibly claim that it was the techno
110.
▲
by
bad_username
7mo ago
Came for Markdown, stayed for a very interesting glimpse into the MS Word file format history.
111.
▲
by
bad_username
7mo ago
This is exactly why Larry Fink makes these statements: to singal the correct virtues and keep himself out of trouble. This is the modus operandi for most people who possess unfairly oversized influence today.
112.
▲
by
bad_username
7mo ago
> so much so that there's a wikipedia page dedicated to the falsehoods Not detracting from the merits of your statement, but Wikipedia is not neutral, it is biased politically/ideologically, so it should not be used as a fair &
113.
▲
by
bad_username
7mo ago
> Almost every male character in The Big Lebowski is, in some way, a failed version of what he thinks he should be. What a fresh, contrarian take! So corageous.
114.
▲
by
bad_username
7mo ago
> There is no world where you input a document lacking clarity and detail and get a coding agent to reliably fill in that missing clarity and detail That is not true, and the proof is that LLMs _can_ reliably generate (relatively small a
115.
▲
by
bad_username
7mo ago
HN is turning into Reddit. It's noticeably worse than it was 5 years ago.
116.
▲
by
bad_username
8mo ago
Sounds like you didn't lose the ability, you lost motivation. Why learn Rust, you say, if an LLM can crank out a Rust app for me, and it will be good enough? LLMs may have removed the critical need for a SW engineer to know details, li
117.
▲
by
bad_username
8mo ago
Given the modern trend of using genderless singular "they", that seems intentional (and off-putting).
118.
▲
by
bad_username
8mo ago
The author's concern of "misgendering" an imaginary person (with ab unambiguously female name) is quite odd.
119.
▲
by
bad_username
8mo ago
> you can remember them, but are you also suggesting to not use generated passwords? You can remember a strong generated password if it's a pass phrase. Better "rememberability" with the same amount of entropy.
120.
▲
by
bad_username
8mo ago
I feel this OpenClaw stuff is a bit like the "crypto" of agentic AI. Promise much, move fast and break things, be shiny and trendy, have a multitude of names, be moderately useful while things go right (and be very useful to malic
More ›