Post is at least semi AI-generated and contains conflicting or misleading bits. It could (and should) have been expressed differently. It conflates changing one flag with permuting a flag at a time in different places, leaving me unable to understand what was actually tried. It has warnings on interpreting the results that make no sense. If flags were permuted, it isn’t possible to list their impact one-dimensionally, then also mixes up addition/removal of flags with permuting its individual value.
I regret giving this article a chance and wasting my time trying to figure out why the “author” was saying what they were saying. Flagged.
Im not a native English speaker so it took me a while longer but once I noticed? Hell no, if it's not worth the time to write or at least edit it, it's not worth the time to read.
And that's even before taking the contradictiona into account.
I'm not really sure about that exactly, and the stream-of-Claude text doesn't lend a lot of confidence in the results, but running this with the command line driver wrapped in a scipy.optimize script seems like a faster and cleaner route to an accurate result.
Much Claude, very 4.7: "Heavy lifting... single most useful thing... the careful claim is narrow... there are three outcomes, not two... honest caveat... honest oddity... honest verdict... honest shape."
It's the weekend on HN, prime time for lost souls, so I'll dare to post that.
If you're going to prompt out an essay, at least take a minute to set up the prompt so that a hint of your personality, or even an invented one, comes through.
If you're still running a MacPro4,1->MacPro"5,1" conversion, be really careful attempting this: one of the few differences is that its bootrom only has a single VS_Cache (i.e. the real MacPro5,1 , manufactured 2010+, has two VS_Cache).
This can lead to a VS_Store corruption == noBOOT sadMAC
It turns out that the essentially no difference thing wasn't true... but fortunately these things are such antiquated beasts that just two people can fix this, via email, by custom "rebuilding your VS_Store firmware" – you'd have to search the forums.
One sad day soon... the "5,1"s won't boot (without pre-emptive maintenance, or an onboard reflow of the many-pinned bootROM chip). Every bootup... one step closer to its eWaste fate.
----
Yes, it uses a lot of power, but my VEGA64 is about as fast as an RTX 5060Ti(8gb) for using Ollama, locally (eight year old design, but it used HBM2 !!).
In wintertimes, this thing has reliably acted as a monero-mining space-heater, albeit quite inefficiently (no: I do NOT have a reversible heatpump; spaceheaters, only == no net-cost).
Truly this is the autoexec.bat/xf86config tuning of our modern age. But how long will it take this time until all this painfully accumulated knowledge is obsolete?
Was already getting lots of AI generated vibes from the article before that, but I’ve seen Claude constantly use that phrase and no real person ever in my life (though I guess some people must have or LLMs wouldn’t have learned it).
It’s the most annoying thing about reading anything LLM related. I genuinely (ugh that’s a term I’ve picked up from using Opus 4.8 every day for work as it constantly says it and now so do I and I can’t stop) want to learn more about the topic but it’s painful to read most posts about it.
The people really into LLMs, surprise!, have a tendency to also LLM generate their writing about LLMs. This would be fine if they used that as a starting point then edited it for brevity (LLMs are consistently overly verbose for some reason, this post was like twice as long as it needed to be, though I guess you could argue the same about my comment haha), correctness, and tone…but they usually don’t.
Then I make it halfway through the article and wonder if I can trust any of it at all.
Interesting results, but the piece is written in an almost repulsive claudglish (or geepeeteeglish). Can’t force myself to read more than a couple sentences.
Post is at least semi AI-generated and contains conflicting or misleading bits. It could (and should) have been expressed differently. It conflates changing one flag with permuting a flag at a time in different places, leaving me unable to understand what was actually tried. It has warnings on interpreting the results that make no sense. If flags were permuted, it isn’t possible to list their impact one-dimensionally, then also mixes up addition/removal of flags with permuting its individual value.
I regret giving this article a chance and wasting my time trying to figure out why the “author” was saying what they were saying. Flagged.
100%
Im not a native English speaker so it took me a while longer but once I noticed? Hell no, if it's not worth the time to write or at least edit it, it's not worth the time to read.
And that's even before taking the contradictiona into account.
I think the submission doesn't belong to HN.
AI detector 96%
Seems like this process itself should be incredibly amenable to automation. Here's 25 parameters, permute over them, perform this test on each loop.
But it sounds like there were other gremlins (like ulimit reverting itself) that would've thwarted automated testing.
Do you feel like there's a point where it would've made sense to automate?
I'm not really sure about that exactly, and the stream-of-Claude text doesn't lend a lot of confidence in the results, but running this with the command line driver wrapped in a scipy.optimize script seems like a faster and cleaner route to an accurate result.
That and I’m not sure why they rebooted the server every time instead of just changing the flags…
I think he did employ automation?
it's just that each run takes a long long time
Much Claude, very 4.7: "Heavy lifting... single most useful thing... the careful claim is narrow... there are three outcomes, not two... honest caveat... honest oddity... honest verdict... honest shape."
It's the weekend on HN, prime time for lost souls, so I'll dare to post that.
If you're going to prompt out an essay, at least take a minute to set up the prompt so that a hint of your personality, or even an invented one, comes through.
If you're still running a MacPro4,1->MacPro"5,1" conversion, be really careful attempting this: one of the few differences is that its bootrom only has a single VS_Cache (i.e. the real MacPro5,1 , manufactured 2010+, has two VS_Cache).
This can lead to a VS_Store corruption == noBOOT sadMAC
It turns out that the essentially no difference thing wasn't true... but fortunately these things are such antiquated beasts that just two people can fix this, via email, by custom "rebuilding your VS_Store firmware" – you'd have to search the forums.
One sad day soon... the "5,1"s won't boot (without pre-emptive maintenance, or an onboard reflow of the many-pinned bootROM chip). Every bootup... one step closer to its eWaste fate.
----
Yes, it uses a lot of power, but my VEGA64 is about as fast as an RTX 5060Ti(8gb) for using Ollama, locally (eight year old design, but it used HBM2 !!).
In wintertimes, this thing has reliably acted as a monero-mining space-heater, albeit quite inefficiently (no: I do NOT have a reversible heatpump; spaceheaters, only == no net-cost).
Truly this is the autoexec.bat/xf86config tuning of our modern age. But how long will it take this time until all this painfully accumulated knowledge is obsolete?
I never want to read the phrase "heavy lifting" again.
Or “belt-and-suspenders”.
Was already getting lots of AI generated vibes from the article before that, but I’ve seen Claude constantly use that phrase and no real person ever in my life (though I guess some people must have or LLMs wouldn’t have learned it).
It’s the most annoying thing about reading anything LLM related. I genuinely (ugh that’s a term I’ve picked up from using Opus 4.8 every day for work as it constantly says it and now so do I and I can’t stop) want to learn more about the topic but it’s painful to read most posts about it.
The people really into LLMs, surprise!, have a tendency to also LLM generate their writing about LLMs. This would be fine if they used that as a starting point then edited it for brevity (LLMs are consistently overly verbose for some reason, this post was like twice as long as it needed to be, though I guess you could argue the same about my comment haha), correctness, and tone…but they usually don’t.
Then I make it halfway through the article and wonder if I can trust any of it at all.
Interesting results, but the piece is written in an almost repulsive claudglish (or geepeeteeglish). Can’t force myself to read more than a couple sentences.
As soon as I saw the phrase “belt-and-suspenders” I immediately thought Claude, as it uses it constantly, but maybe so does ChatGPT.
“worth sitting with ” is in the way to overused scale :(
Maybe the author had something interesting to say! Maybe they don't!
But I won't be the one to tell you, because I sure as fuck am not going to wade through overly verbose, tired linkedin-tier slopisms to find out.
The "author" being distinctly LLM.
Can HN please ban AI generated slop ?
Please...
but why would you restart the whole machine?
(otherwise, congrats for discovering the scientific method!)