Hacker Newsnew | past | comments | ask | show | jobs | submit | dnautics's commentslogin

> the chinese are definitely going to pull ahead

Are they? Please don't forget at least 50M Chinese in abject poverty and a demographic crisis barreling towards them that is not avoidable unless they develop cloning vats that can rapidly age a clone to productive adulthood in a compressed timeframe.


the heartbreaking truth is that those 50m people do not matter or at the very least the government is not interested in them.

"do not matter". I mean idk, some of them burn down factories when they don't get paid.

nothing a genocide or two can't solve (based on a real story)

the governments have grown too powerful to be overthrown by people, and yes I do realize that the military itself is made out of people (for now), but in a way they are brainwashed? to follow orders from above.


that's just perl with a new coat of paint

Perl 5 does not have types, as far as I know.

I’m taking the GP seriously instead of dismissing it. Raku looks like more fun than nushell tbh.



    print 42 + 99;          # 141
    print &print.file ;     # ...src/core.c/io_operators.rakumod
    print &infix:<+>.file;  # ...src/core.c/Numeric.rakumod
    print ?CORE::<&print>;  # True 
I barely understood these four example lines.

It has. They are just poorly supported and disliked.

it is so much slower to start than perl, that you don't want to run it for quick one-off tasks

well waiting for Raku (formerly perl6) to be built was like watching paint dry

They should have named it Godot. ;) But now that name is taken.

Lead paint with asbestos sprinkles and radium racing stripes! ;)

you understand that stack allocation can OOM too?

Can C gracefully recover from running out of stack space?

Depends on your definition of graceful but the C standard doesn't preclude handling it and there's POSIX interfaces such as sigaltstack / sigsetjmp etc that fit and indeed some code like language runtimes use this to react to stack exhaustion (having first set up guard pages etc).

Or in short: no, C is no better than Rust at gracefully recovering from stack overflow, either in theory or in practice.

not enforced for any given implementation is hardly "unknown". presumably the tin comes with a label saying what's inside

as someone building an analyzer for zig, if you sent something like this through the system im building, it would get handled by Zig's optional type tagging; but lets say you did "free" (so your result is half freed, half not freed): it would get rejected because both branches produce inconsistent type refinements, you don't have to solve the halting problem, just analyze all possible branches and seek convergence.

Rust is the poster child for these complaints, but this is a great example of "the language rejects a valid program". Not all things that can be expressed in C are good ideas!

This is "valid" C, but I wholly support checking tools that reject it.


exactly! "guaranteeing the safety of C" sir what did you think that meant, sprinkling magic fairy dust to make it work!!?

i made a quip and realized that's not a bad description of what fil-c does

Are you implying that Fil-C has this sort of reaction to people confused about why it does certain things in the name of safety, or are you saying Fil-C is just sprinkling magic fairy dust on C and declaring it safe?

i like fil-c, so i would say "fil c sprinkles magic fairy dust on C and makes it safe (at the cost of perf and elevated risk of crashing)

> just analyze all possible branches and seek convergence.

This sounds like a very simple form of abstract interpretation, how do you handle the issues it generally brings?

For example if after one branch you don't converge, but after two you do, do you accept that? What if this requires remembering the relationship between variables? How do you represent these relationships?

Historically this has been a tradeoff between representing the state space with high or perfect precision, which however can require an exponential amount of memory/calculations, or approximate them with an abstract domain, which however tend to lose precision after performing certain operations.

Add dynamically sized values (e.g. arrays) and loops/recursion and now you also need to simulate a possibly unbounded number of iterations.


> For example if after one branch you don't converge, but after two you do, do you accept that?

you should refactor so that it's representable.

> Add dynamically sized values (e.g. arrays) and loops/recursion and now you also need to simulate a possibly unbounded number of iterations.

regions are hard. You kinda have to reject regions that are not uniform. loops you can find a fixpoint for.


eh, germany and japan seemed to go okay, grenada too. korea kind of a mixed bag (it took decades for it to not suck)

Korea, by what metric? South Korea was through the 50 poorer than North Korea, North Korea was considered the roaring growth economy, huge success of planning and leadership.

Park Chung Hee took a country that could not be a functional democracy, provided leadership and put it onto the path of economic success. Iirc, the reduction in poverty through that period is the fastest in human history (when you consider that China, that is an incredible statement).

I think people (still) assume both that democracy is superior economically for every situation and that people who don't have any food care about being unable to vote...neither of these things is obviously true. Indeed, in the latter case, we now have a good test case of poor countries adopting democracy early and they have generally not been successful as power rotates between various quasi-dictators who give massive handouts to the poor to retain power (without doing anything actually useful).


The choice should be free though - everyone should be able to opt out. Restricting people to leave the country is a major red flag that something is going in the wrong direction.

Crediting the US for Germany's post-WWII transition is a bit of a stretch. There were quite a few more players involved.

Considering the Soviet strategy of stripping assets from the East and the fact that Britain and France were broke and in shambles - yes, the Marshall plan deserves great credit. To this day East German states remain the poorest in the nation.

To be fair they had kind of started implementing the Morgenthau Plan until they realized that maybe it wasn't the best idea (and the British played a significant role in convincing the US government about that...)

1945 to ~1947 were very rough in Germany even in the allied occupation zones (and that was at least partially an outcome of a conscious decision by the allies to not allow German industry to recover)


> grenada too

Grenada is something of a joke in this context - the entire thing came about because the communist government fell apart and started fighting internally, so it's pretty likely the regime would have shortly collapsed with or without the invasion


The scale of investment and commitment was orders of magnitude larger, as was the utter devastation inflicted for years before hand. Incomparable.

unlikely. trump didnt held obama accountable for all sorts of crazy things that happened during his administration (bombing libya, drone striking a us citizen minor, using USAID to mount a fake vaccination campaign for DNA surveillance in pakistan e.g.). why would the next administration hold trump accountable?

The Biden administration was prosecuting Trump though. They didn’t complete the prosecutions because Trump’s strategy to avoid accountability was to be reelected and then shut down the investigations, and that worked. But the fact he was indicted by Jack Smith who very likely could have convicted him goes to show lack of accountability is not for lack of trying.

Its very much for lack of trying. They had 4 years, we got no epstein files and they slow walked prosecutions to happen during the election, thinking it would help them. It didn't work, here we are.

It’s clear you didn’t follow these cases if your opinion is the SC slow walked them to enhance Democrats’ electoral out look. They secured multiple indictments and were heading to trial, which they were likely to win. Delays were caused by Trump appointed Judge Cannon and Trump appointed SCOTUS justices.

Securing indictments and going to trial is an instance of actually trying. So you really can’t say they didn’t try, because that is factually false. It’s true they could have done more, but they didn’t do nothing as others are saying.


I'm not a lawyer, and I didn't follow every motion, you're right. Still, in my book, fast walking would have meant moving faster. Venue shop if you have to. Release/declassify documents to make the bad guys look bad. There's lots of "improper" stuff they could have done and are currently getting owned by.

I'm not a lawyer either but I did follow the cases closely. My opinion is that Merrick Garland did a disservice to the country by not appointing a SC immediately, but beyond that Jack Smith moved with lightning speed in prosecuting the cases. Moreover, Congress did make the bad guys look bad -- they held a whole summer's worth of hearings where they prosecuted the case in public, offering plenty evidence. And I encourage you also to look at how it was the Supreme Court who slow walked their decisions, which ultimately benefitted Trump in obscene ways. You can't venue shop SCOTUS.

One thing about prosecuting a former POTUS for the first time is it has to survive the test of time. You can't behave like them if you want the prosecution to be legitimate, because they are lawless. But it was the failure of voters to do their due diligence to not elect a felon who bear the ultimate blame, as they are the final check. Now we bear the consequences. But again, not for lack of trying.


It's late where I am so I don't have a well-reasoned response, just wanted to say I understand what you're saying. It sucks, given what the current admin is getting away with, but I understand it.

i would feel better about that if the biden administration also prosecuted obama. they didn't. besides trump I (nor biden) didnt do any new foreign adventures AFAICT. we had a blissful 8 years of waning US imperialism

It's unclear if most if not all of those things you were actually crimes legally (regardless of how morally and ethically reprehensive they might have been). Regardless there was an established precedent for what Obama was doing. Not so much for the crimes Trump was being accused..

Abdulrahman al-Awlaki was a 16-year-old United States citizen who was killed by a U.S. drone strike in Yemen, a country with which the United States was not at war with.

Please let me know what was the established precedent for allowing extrajudicial assassination of American citizens is.

Edit add:

He was a boy who was still searching for his father when his father was killed, and who, on the night he himself was killed, was saying goodbye to the second cousin with whom he'd lived while on his search, and the friends he'd made. He was a boy among boys, then; a boy among boys eating dinner by an open fire along the side of a road when an American drone came out of the sky and fired the missiles that killed them all.

A 16-year-old American boy accused of no crimes was killed in American drone attack

https://www.esquire.com/news-politics/news/a14796/abdulrahma...

So please, I would love to see the precedent.


pretty sure when obama murdered Abdulrahman al-Awlaki (nb: not talking about the more famous Anwar) that was unprecedented. Trump later murdered Abdulrahman's sister, but at that point, it was "precedented" by obama.

He did prosecute his political opponents like Bolton though for doing exactly what Trump did just on a likely several magnitudes smaller scale...

All this fuckery date from at least bush 2nd. Election mess, with heavy involvement of his brother the governor despite promises to revise, crowds attacking poll workers, war crimes, putting incompetent friends at the head of agencies (remember FEMA response to Katrina? Or the initial response to the subprime crisis?), attacks on science programs and schools, and the use of executive orders to bypass congress. Obama was so tame compared to Bush2.

I don't think this is correct. For inference, the bottleneck is memory bandwidth, so if you can hook up an FPGA with better memory, it has an outside shot at beating GPUs, at least in the short term.

I mean, I have worked with FPGAs that outperform H200s in Llama3-class models a while and a half ago.


Show me a single FPGA that can outperform a B200 at matrix multiplication (or even come close) at any usable precision.

B200 can do 10 peta ops at fp8, theoretically.

I do agree memory bandwidth is also a problem for most FPGA setups, but xilinx ships HBM with some skus and they are not competitive at inference as far as I know.


Said GPUs spend half the time just waiting for memory.

Yep, but they are still 50x faster than any fpga.

probably not B200 level but better than you might expect:

https://www.positron.ai/

i believe a B200 is ~3x the H200 at llama-3, so that puts the FPGAs at around 60% the speed of B200s?


I wouldn't trust any benchmarks on the vendors site. Microsoft went down this path for years with FPGAs and wrote off the entire effort.

ok? i worked on those devices, those numbers are real. theres a reason why they compare to h200 and not b200

> I have worked with FPGAs that outperform H200s in Llama3-class models a while and a half ago


I'd like to know more. I expect these systems are 8xvh1782. Is that true? What's the theoretical math throughput - my expectation is that it isn't very high per chip. How is performance in the prefill stage when inference is actually math limited?

i was a software guy, sorry, but those token rates are correct and what was flowing through my software.

i believe there was a special deal on super special fpgas. there were dsps involved.


I don't know if they'll reach $3B, but at least one company is using FPGA transformers (that perform well) to get revenue in before going to ASIC transformers:

https://www.positron.ai/


i hate python but if your bottleneck is that sqlite query, optimizing a handful of addition operations is a wash. thats why you need to at least have a feel for these tables

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: