yzydserd

If you enjoyed the article, consider listening to the Fall of civilisations podcast. Episode 12 spends 3 hours on the whole chilling saga.

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

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 non-malicious, i.e. other sites may not be able to see as much.

> 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.

Which part was ironic?

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 new DNA evidence used in actual innocence decisions. Too often, a jury convicts a likely suspect on weak circumstantial evidence. Just as with Touch DNA, Air DNA will create new problems with avoiding contamination.

The path toward Air DNA has been known for years [0]. Wouldn’t be surprised to learn crime scene investigators have been sampling and storing air in high profile cases ready for the tech to catch up.

[0] https://www.science.org/content/article/dna-pulled-thin-air-...

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”.

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.

Sadly, it has a small context window.

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 rest of the 64KB TCM. There’s just a 4KB cache. So the limiting factor is often memory access not cpu. The 400 x 240 1 bit Sharp Memory LCD has to be carefully addressed in horizontal strips of 32 bit words with weird endianness if you want high performance. It’s very possible to work at the maximum 50fps per full frame and do per pixel work and prep for blit in sub 15 percent of CPU. Almost all devs aim for and achieve 30fps. You can have effective 200fps if you like since not all rows need to be refreshed per frame. An interesting situation is the device had an earlier Rev A before the Rev B that 90% of users have, and with Rev A up to 30-50% slower you have to develop for the base device specs if you want to be inclusive. It makes you wonder about Rev C decisions. All that said, most games are written in Lua which runs 50x slower than compiled C on the device, yet many very fun games have been released. You can compile ARM intrinsics/asm but it’s a rare situation if you can beat the compiler. Working within all these constraints is part of the charm of developing for it. It necessitates a discipline of writing tight benchmarked code, even inspecting compiled instructions, if you want to push it.

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 presidential terms are “long tones”. Weird memory functions.

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.

Its due to carry forward of net operating losses from previous years, stock compensation deductions, and other depreciations and deductions.

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.

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 keep up with you at thinking / typing speed, whereas o1-pro can take several minutes. Just a quick view after playing for an hour.

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.co.uk/politics/articles/51484-how-do-britons-...

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.

Nice idea.

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.

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 understand the jury system and prosecutorial process?

Also, I have watched a bunch of series and I have not once noticed anyone click the obvious “hack”: the last person to enter the breakfast room after a kill night is almost always a faithful, because of the TV cliffhanger of viewers hanging on which of two faithful survived. It’s the best truth signal the game gives and I’ve never seen a player mention it. Maybe TV edits the knowledge out.

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.

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 gender, and there isn’t at the time of comment.

I called someone out recently for the same, incurring -4 karma, and it turned out the gender pronoun was wrong: https://news.ycombinator.com/item?id=42514127#42520946

How do you know it’s a male?

I’m being downvoted for each comment I make on the topic but anyway …

I found a source article [0] about how the Bureau adjusts the cost for features with a “hedonic” score. So yes, a new feature (a Netflix app? AI? Voice recognition? Soundbar?) is counted as an improvement with value and that in turn affects a prior.

On another BLS page: “if a television in the CPI is replaced by one with a larger screen and higher price, the BLS can make an adjustment to the price difference by estimating what the old television would have cost had it had the larger screen size.”

I bought a 42” mid range TV in 2012 for 900, a 55” mid range in 2017 for 900, and a 55” mid range in 2024 for £1200. Although this is a 30% or so discount (in inflation adjusted terms) the BLS can call it 80% because it has Smart Features like OLED and internet connectivity and so forth.

[0] https://www.bls.gov/cpi/quality-adjustment/questions-and-ans...

That does sound right. But imagine we're plotting the cost of a laptop from 2015 to 2025. We're getting say 4x the memory and 4x better speed (if not more), but would we show a graph that indicates laptops are 4x cheaper in 2025 than 2015? I wouldn't.

Right, but imagine that with cars not TVs. To show the cost change of a car from 2000 to 2020 I wouldn't show the cost of a 2000-quality car in 2020. I'd show the actual median cost of a car in 2020.

I'm not a big fan of the price graph in the article.

Are TVs really 90% cheaper since 2020 than they were in 2000? The product seems to me to be one of those that always costs the same, you just get more for a similar amount of money as time progresses (like laptops).

What does it mean for the average hourly wage line to show a 100% increase? Like it shows clothing to be a constant cost. But that's before a 100% increase in hourly wages, so has effective clothing cost halved? And food and beverage tracks the wage increase, so is it the same effective cost?

And you get artefacts like the 90% drop in TV price looking to be the same scale as an 80% increase in food and beverage cost. I can buy 10 times as many TVs, but half as much food.

Why is is More expensive vs more affordable? Seems odd to mix terms of affordable and expense when the graph isn't corrected for earnings.

I suppose it may work for some.

Or at least defenestrate the mouse.

0-25: learning

25-50: doing

50-75: enjoying

75-100: leaving

Source code: https://github.com/grzegorz-kraszewski/untangle