john-h-k

23 ms·

121.
by john-h-k 1y ago
The legal system is when you get to shoot at things that break the law
122.
by john-h-k 1y ago
x86 and modern ARM64 has instructions for this, but original ARM and RISC approaches are literally a hardware-assisted polling loop. Unsure what guarantees they make, I shall look. Definitely a good clarification though yeah, important
123.
by john-h-k 1y ago
note: since learnt that B port is just port 11 in all the intel docs, uops.info just hexifies them to keep ports single-char
124.
by john-h-k 1y ago
> The important thing to remember is that each of these cannot be split into separate instructions. Nitpick, but they absolutely can be split into several instructions, and this is the most common way it’s implemented on RISClike process
125.
by john-h-k 1y ago
This isn't just due to the actual dependencies of flag instructions at hardware level (although likely be a factor), it also majorly affects code layout. On Arm64 for example, you can make a comparison, do other operations, and then co
126.
by john-h-k 1y ago
uops.info has latency for both (Alder Lake) at 1 cycle but throughput (lower is better) * for add is 0.20 (ie 5 per cycle) * for adc is 0.50 (ie 2 per cycle) so it does seem correct. This seems to be a consequence of `add` being available o
127.
by john-h-k 1y ago
> This article, while its tone aged poorly with the relevant election results, While I get your point, it’s critical to recognise that betting 90% on a six being rolled doesn’t make you correct when a six is rolled. You can believe polym
128.
by john-h-k 1y ago
Yes, I am wrong it seems. I am unsure where I got the high<->secondary mixup from, I'm sure I saw something talking about high explosive and defining it strictly as "needs supersonic shock wave to initiate", but have ju
129.
by john-h-k 1y ago
I’ll happily admit if I was asked to do this I’d run away screaming. That being said, once it’s failed detonation (and you’ve cut off any possible signals to the detonator), wouldn’t you roughly expect it to be as dangerous as transporting
130.
by john-h-k 1y ago
next black mirror episode will be someone screaming for help as they are trapped in the bathroom while their anti-brown-noise device drowns out their pleas and they are left to die in there
131.
by john-h-k 1y ago
Right. But given all pairs of mainstream LLM combinations, it seems a model is more likely to say “yes I am X” when it is X than when it isn’t X, even if it still has a high chance of being wrong. Which means you should (as a bayesian actor
132.
by john-h-k 1y ago
> General point: it's impossible to prove anything based on an LLM's response since it's impossible to distinguish a true LLM statement from a false one. This seems true but sort of vacuous. Obviously an arbitrary statemen
133.
by john-h-k 1y ago
https://upload.wikimedia.org/wikipedia/commons/thumb/b/b2/Su...
134.
by john-h-k 1y ago
Wow not even a verbal slip up I literally read Australia as Austria. Equally impressed and horrified with myself
135.
by john-h-k 1y ago
> a next token predictor can figure it out Describing LLMs as "next token predictors" is disingenuous and wrong
136.
by john-h-k 1y ago
> We haven't solved interpretability and we don't actually know how these things work But right above this you made a statement about how they work. You can’t claim we know how they work to support your opinion, and then claim
137.
by john-h-k 1y ago
Don’t OBB Railjet trains travel at 230km/h?
138.
by john-h-k 1y ago
Very few compilers actually use vertex coloring for register allocation
139.
by john-h-k 1y ago
> At best, this screams, “you’re doing it wrong.” If you’re doing it wrong to a meaningful extent you won’t be able to get insurance or it will be very expensive
140.
by john-h-k 1y ago
This feels like a pretty far fetched straw man. If someone invented medical malpractice insurance yesterday you could use the exact same argument. More generally I think “if something is bad, we should not be able to insure it because the
141.
by john-h-k 1y ago
But you can load any byte by loading 8 bytes and shift (v cheap)
142.
by john-h-k 1y ago
Obligatory link about AI water/energy use https://open.substack.com/pub/andymasley/p/individual-ai-use...
143.
by john-h-k 1y ago
> Yes, deep learning models only interpolate What do you mean by this? I don’t think the understanding of LLMs is sufficient to make this claim
144.
by john-h-k 1y ago
To be fair, only 3 posts within "possible LLM usage" timeframe. Also I don't think using LLM to comment == bot. More curious about the motivation behaviour such behaviour, if it is occurring
145.
by john-h-k 1y ago
I find this comment odd but not at all discriminatory and will happily inform that yes, I am white British. I can see the point you’re going for but I do think it’s more complex than that. I don’t think LLMs are equivalent to another human
146.
by john-h-k 1y ago
Was this written by an LLM? Not accusative but something about the vibe strongly suggests it
147.
by john-h-k 1y ago
Ah, interesting to know. I guess it is only the coreutils stuff where that is a concern
148.
by john-h-k 1y ago
The other big thing iirc is they’re all MIT licensed rather than GPL(et variants) licensed
149.
by john-h-k 1y ago
Sorry, yes, that was a major misunderstanding of the previous comment by me. I am utterly wrong!
150.
by john-h-k 1y ago
[remarkably incorrect answer misunderstanding the above comments]

More ›