Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
yzydserd
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
8 ms
·
91.
▲
by
yzydserd
2y ago
Seems fair. They charge for online collaboration through NSO. This is another online feature.
92.
▲
by
yzydserd
2y ago
While the framing has been reciprocal tariffs, the WH has published the formula and it’s essentially based on reciprocal trade surpluses, not tariffs. If one country sells more to the US than the US sells back then that’s seen as bad. Even
93.
▲
by
yzydserd
2y ago
And consumers will pay the $12. It won’t make it viable to make the bike in the US for < $40.
94.
▲
by
yzydserd
2y ago
> Family Plan … We strive to provide a search engine that prioritizes the well-being of your loved ones, particularly the most vulnerable ones like children, by offering an ad-free and safe browsing experience. We offer two different gro
95.
▲
by
yzydserd
2y ago
What’s the old adage, is it “I don’t know how WW3 will end, but WW4 will end with rock paper scissors”? Right now, my money’s on scissors.
96.
▲
by
yzydserd
2y ago
If you enjoyed the article, consider listening to the Fall of civilisations podcast. Episode 12 spends 3 hours on the whole chilling saga.
97.
▲
by
yzydserd
2y ago
I didn't read every word of the PDF, but I didn't see a definition of coastline derivation. See the Coastline Paradox https://en.wikipedia.org/wiki/Coastline_paradox
98.
▲
Curiosity rover finds long-chain alkanes on Mars
(theguardian.com)
3 points
by
yzydserd
2y ago
|
1 comments
99.
▲
by
yzydserd
2y ago
Not the poster, but for me MacOS Orion (without stop the madness or little snitch etc) or anything else scores 16.07 bits and 98% ad block. I had a look at Kagi official discord and Vlad (HN: @freediver) says they let the eff test run as no
100.
▲
by
yzydserd
2y ago
> Much improved Kobo support. calibre can now natively edit, view and convert KEPUB format files used by the Kobo. It also automatically converts EPUB to KEPUB when sending books to Kobo devices. Great to see improved Kobo support.
101.
▲
by
yzydserd
2y ago
Which part was ironic?
102.
▲
by
yzydserd
2y ago
It’s possibly not the GP’s point, but in general the more DNA that is available the better from a defense perspective. Many wrongful convictions involve a lack of physical evidence. Recent advances like “touch dna” and “m-vac” have led to n
103.
▲
by
yzydserd
2y ago
Many prompters don’t practice hygienic reversal by rolling the outcome of a conversation back up the thread as input. If 10 comments are spent coming to some clean code, go back up the thread and include it as the status “back then”.
104.
▲
by
yzydserd
2y ago
Here is Simonw experimenting with ChatGPT and C a year ago: https://news.ycombinator.com/item?id=39801938 I find ChatGPT and Claude really quite good at C.
105.
▲
by
yzydserd
2y ago
Sadly, it has a small context window.
106.
▲
by
yzydserd
2y ago
It has a Cortex M7 180 MHz MCU. No real SIMD. Reasonable memory at 16MB but it is slow to r/w at about 30MB/s. DTCM 30x faster but you don’t get much to address, e.g. the stack is just 10KB. Framebuffer 12KB. System reserves the r
107.
▲
by
yzydserd
2y ago
For me, I employ presidential terms as memorable 4 year labels, much the same as Olympiads (Beijing 2008, Sydney 2000, Barcelona 1992). It’s got zero to do with (geo) political influence. The Olympiads work as point in time “beats”, while p
108.
▲
Surface stable fractal dither on Playdate
(aras-p.info)
1 points
by
yzydserd
2y ago
|
0 comments
109.
▲
by
yzydserd
2y ago
How do today’s LLM’s like Gemini compare with the Document Understanding services google/aws/azure have offered for a few years, particularly when dealing with known forms? I think Google’s is Document AI.
110.
▲
by
yzydserd
2y ago
Its due to carry forward of net operating losses from previous years, stock compensation deductions, and other depreciations and deductions.
111.
▲
by
yzydserd
2y ago
Thanks for the Lander pointer. Virus blew my mind when I first saw it, those decades ago. Now to restrain myself from hours in the source code.
112.
▲
by
yzydserd
2y ago
A few quick tasks look to me like o3-mini-high is 4-10x faster for 80% of the quality. It gives very good and sufficient fast reasoning about coding tasks, but I think I'd ask o1-pro to do the task ie provide the code. o3-mini-high can
113.
▲
by
yzydserd
2y ago
The article seems to hyperlink to a 2020 poll of polls for what people are “now saying”. Here is a current poll. “11% of Britons say Brexit has been a success, with more Leavers saying it has gone badly than well” https://yougov.
114.
▲
by
yzydserd
2y ago
That’s not my recollection at the time. It was always 16-bit. The Archimedes was often portrayed with the USP of being 32-bit at the time.
115.
▲
by
yzydserd
2y ago
Nice idea.
116.
▲
by
yzydserd
2y ago
Yes. Another disincentive for finding a traitor in the first half of the game is that they just get replaced, and you get a target on your back. A good playing strategy is to be just vocal enough, with some open opinions.
117.
▲
by
yzydserd
2y ago
The worrying observation I make from Traitors is how easily a group of about a dozen people can become so sure of guilt on so little evidence, and time after time of being knowingly wrong. What if anything can we learn from this to understa
118.
▲
by
yzydserd
2y ago
I think there is a distinction between “instructions”, “guidance” and “knowledge/context”. I tend to provide o1 pro with a LOT of knowledge/context, a simple instruction, and no guidance. I think TFA is advocating same.
119.
▲
by
yzydserd
2y ago
Yes. I simply hope the commenter reflects and doesn’t use a gender next time if there is no evidence of gender. That’s simply it. In 2025 I don’t feel it is appropriate for a commenter to use the pronoun “him” when there is no evidence of g
120.
▲
by
yzydserd
2y ago
How do you know it’s a male?
More ›