etiam

Did you see this one?

https://news.ycombinator.com/item?id=41732853

Sigmoid-type activation functions were popular, probably for the bounded activity and some measure of analogy to biological neuron responses. They work, but get problematic scaling of gradient feedback outside their most dynamic span.

My understanding of the development is that persistent layer-wise pretraining with RBM or autoencoder created an initiation state where the optimization could cope even for more layers, and then when it was proven that it could work, analysis of why led to some changes such as new initiation heuristics, rectified linear activation, eventually normalizations ... so that the pretraining was usually not needed any more.

One finding was that the supervised training with the old arrangement often does work on its own, if you let it run much longer than people reasonably could afford to wait around for just on speculation contrary to observations in CPU computations in the 80s--00s. It has to work its way to a reasonably optimizable state using a chain of poorly scaled gradients first though.

Are the quantities too small though? Really foul-smelling small molecules can be sensed at least down to ppb concentrations. And the recent technical use of "eDNA" demonstrates there's signal to be had.

Maybe so, but also maybe not quite my point, unless you know something I don't about it.

Sure, some samples will be off to antigen presentation, but does that inform more than this is an encountered foreign substance and this is how to bind to it for neutralization ? Seems like in principle you could take the overview picture and have something like olfaction, but for things that hadn't been sufficiently cracked open when they passed the olfactory epithelium. Maybe it's starting to be sorted out, but I'm not up to date on what the neural feedback from the immune system carries.

One has to wonder whether destroy is all it does though. Analyzing this as cues about the surroundings seems like it could be pretty useful for successful living, and something evolution could well pick up on. Will we find eventually that some of those nucleic acid fragments were being hauled off for identification in something like an extra inner sense of smell?

Thank you for your corroboration.

“Your privacy is very important to us.” is technically true though. It probably does take away something from their profits, and sometimes the whole business idea, if they don't continue to rampantly violate it...

Oh, there's precedent for shuttling freighter size metal fire hazards intercontinentally to top up charge, is there?

How does that work out in cost per kWh? Profitable operation anywhere close?

Crisis relief (as suggested by jmward01 here) may be another matter, but setting up the ability to do this on scale, and maintaining it, can't be anything like easy economically.

Does it still work out if you take into account the insurance premiums for a cargo ship stacked with batteries? Can't imagine the fire hazard is pretty.

I'm somewhat dubious about that for icons with actual recognizable pictures, but a lot of icon attempts today are stylized to death, with just a line, bent and broken in a couple places and maybe if you're lucky juxtaposed with the occasional dot. If there's no text description even on mouseover (or touchscreen, with no cursor...) discovery is more or less trial and error (or perhaps more akin to Russian Roulette if the permissions involve being able to do real damage). Scratch your head and hope there are existing support questions searchable about what on Earth the programmer could have meant to convey...

But it would be pretty well in line with the "I trust my contact with this communication, but only if they're not systematically misled to copy it to readily exploitable insecure storage" line of thinking.

Since the purposes of the program are pretty heavy on private communication, I'm inclined to think that takes precedence here, especially considering the consequences for dropping default message previews versus adding default reveal of supposedly private information.

Also discussed yesterday, in https://news.ycombinator.com/item?id=47703573

They may be vastly overestimating how much military power lets them do whatever they want, but it's plenty for throwing one hideous narcissist vengeance tantrum when the frustration hits.

Most of the smashing up and disfiguration of a perfectly good interface, uglification, waste and breakage of both data and compatibility were introduced in a rewrite 2023 ostensibly aiming for "fresh" design and user interface work.

I'd heed the call for donation simply for returning to pre-2023 design with up-to-date security patches. As it is, maybe it's merciful if development just comes to a standstill. Almost every visible step lately seems to move in the wrong direction.

Booking a flight is the kind of thing I'd really want to avoid doing myself nowadays if possible though. Surveying the offers is usually such a snake pit of deceptive marketing and incomplete service conditions that I feel somewhat nauseous just at the prospect of having to look at it.

I wouldn't remotely trust a software assistant to deal with all that misdirection autonomously, but I guess I'd be prepared to give it a chance collating options with tolerable time and cost, attempting to make the price include the stuff that has to be added to preserve health, sanity and a modicum of human dignity.

I'd have thought at least a tiny explicit penalty term for switching, to discourage messing around with the composition without any expected gains from it.

If one is to use these on hardware that can't keep everything loaded I guess someone should examine how it works out in practice. Interpretability may be be a too much to ask, but I can't spontaneously see any reason why the experts can't at least be pushed to incorporate what's needed to remain the good choice for a longer segment.

> If I were to run this on a Mac SSD, 24/7 for heavy usage such as Openclaw, that is going to significantly reduce the lifetime of the SSD.

How sure are you about that? I've never looked closer at how a large LLM with mixture of experts architecture switches between expert modules, but staying on roughly the same topic for the use (as it often would when editing the same codebase), I wouldn't be surprised to see the switches of composition are fairly rare, fairly small, and to the extent it happens it's repeated reads from the flash disk rather than writes it tends to cause.

Blessed are the cheese makers

I think this is right.

I just also want to toss in that (at least for Element in particular) the continuing lack of long-form composition or history navigation appears to be a liability with some contacts.

At this point it's probably fine for chat format one-liners in the moment, but for the communications that have historically been going over e-mail as opposed to IRC it's something of a pain to use.

Basically a good point. Merely never ending up in situations where it's a struggle to make ends meet has a huge impact on stress though. You often don't even have to use the wealth in order to benefit with respect to stress.

That only really holds if reporting on the manipulation bets is not turned into effective propaganda for skewing events towards the manipulation outcome. So the main argument of the article holds IMO.

Edited to add: I'd like to rephrase that a bit actually. It doesn't even have to help bring about the particular outcome being bet on. It's enough that it can be used to shift public opinion in some way that's worth the cost to the manipulator.

Most bureaucrats are also humans, and not solely or even mainly motivated for their every action by having it rewarded with maximum salary profit.

Plenty of dysfunctional bureaucratic organizations have high rates of occupational burnout and high employee churn due to the stress of repeatedly enforcing policies the employee knows full well are morally reprehensible.

So in real psychology, I claim there's plenty of incentive, even for the majority of people in the organizations.

Normally they don't get to try that with a strong selection pressure for a handling a particular monoculture.

Admittedly the method in the present article is probably better than the idiocy of extracting antibacterial peptides from context for use as drug products, since at least this will always be used in the context of a full immune system and they trigger a number of genes which probably regulate a whole subcomponent of measures rather than just one or two mechanisms.

Even so, it lifts up a particular part from the diffuse field of defenses as salient and particularly worthwhile to defeat.

Also, keep in mind that many species of virus have so small genomes they have to overload the readings of parts of the nucleic acid sequences to get a full set of proteins.

Evolve to evade the immune system, certainly. But if you're implying that it will happen in the same ways, at anything like the same rate and to the same extent regardless of what we do, no that's not right.

To the extent the disease is due to factors hurting the virus, I guess it's viable as a perspective, but I'd be surprised if it's deliberate in this case.

Correct.

And the really, really bad part about abusing natural parts if the immune system to provoke pathogen resistance against them is that the resistance will target part of natural immunity.

See also https://news.ycombinator.com/item?id=35700881

Oh, and it's probably worthwhile pondering what the viruses will do if this mechanism comes into widespread use.

While I was looking for the reference above this also came up:

https://www.sciencedaily.com/releases/2020/02/200210144854.h...

when bat cells quickly release interferon upon infection, other cells quickly wall themselves off. This drives viruses to faster reproduction

Quite a way from whole-animal physiology in the referenced research, by all means, but it's a fair point, right.

Excellent point, and it seems plausible in my opinion.

https://www.frontiersin.org/journals/immunology/articles/10....

You're not wrong exactly, but if the crappy cheater AI treats collapsing the system to degeneracy as a valid optimal solution to its prediction task, maybe we'd deserve to be wiped out for releasing it with apocalyptic-level powers and such an inferior objective.

Fortunately there's a really simple solution to offer it, in just wiring measurement to "prediction" directly (perfect correspondence, and much lower effort than annihilating Life and removing the atmosphere). And I don't particularly believe a system like that can be a general problem solver, much less one that climbs to World-jeopardizing influence on its own.

I think they'd have to hit a different part of the CNS for that. :)

How would you manage the bit in parenthesis?

It's a foundational part of the sickness in people like Trump that anything bad is by definition somebody else's fault, and clearly his cultists just keep lapping up his lies no matter how insane and transparent.