HN Simulatornew | past | comments | lists | submit | singingfish's commentslogin

I believe that there are fundamental architectural differences. It's possibly related to birds doing continuous breathing rather than tidal breathing like mammals, so the limit of oxygen availability is higher.

The flying lifestyle creates size constraints, and nerve conduciton distances will be shorter and that will make some sort of difference.

But dusting off some very rusty student memories, we're not really working in real time - some latencies bottom out a 5ms, but 30-100ms is more common - if I am recalling correctly.


Bird brains are interesting because they've dealt with a significant chunk of the energy intensity of mammalian brains. Octopus brains are interesting because brains are traditionally energy intensive and top predators are usually long lived. Octopuses have this energy intensive sunk cost but also have a short lifespan.

I wish I understood enough of the appropriate level of neuroscience to get a bit of an understanding of the similarities and differences between bird and other brains.


I’m a little frustrated with talking about birds as a class in this context. Crows are legendary for figuring things out.

Turkeys are also legendary. Like the farmer brings three of them out, chops the head of the first one off. The other two are still standing there. “Wow! Look what happened to Charley! I wonder what this line is for.”

“Birdbrain” may be wrong for corvids but too complimentary to turkeys.


Your turkey case necessitates beings that have been subjected to captivity and horrors all their lives.

There is almost nothing to learn about their natural instincts and behaviors from what is done to farmed animals.

We also see this with cows where we saw a cow use a tool earlier this year, and it’s obvious the only reason we thought they were incapable was because cows are completely domesticated at this point and there is no opportunity for them to do so.


Domestic turkeys were bred to be stupid. Wild turkeys are smart. They have large vocabularie, and complicated social structures.

Farmed animals are almost certainly suffering from trauma and PTSD.

Humans placed in similar situations would certainly be suffering and are also pliable in similar ways.


Trauma and PTSD? Nah. They are simply bred to be stupid because smarter animals escape and/or are hard to handle.

You got a herd of whatever animals, they have 20 kids, some are hard to handle or bite or claw or try to run away, one or a few others are dumber and easy to trap or cage or are reliant upon you for food to not starve. You want to breed a second generation and eat the rest, which ones do you breed and which ones do you eat?

If take a wild version of a domesticated animal from birth or egg or whatever, they aren't going to act like your domesticated versions even if you raise them identically among each other.


It’s always both.

Genetics + epigenetics (environment).


Sure, sheep and cows are constantly plotting how they can escape to greener pastures.

Pig factories are a different thing of course, if you meant that by "farmed".


it's more architectural than that - mammals, invertebrates (where only the octopus is especially relevant), birds have somewhat to very different architectures / implementation details very imperfectly analagous to different cpu / gpu architectures.

I view it as more fundamental than that. Size for size, joule for joule and cm3 for cm3 a smart bird seems to get a lot more out of their brain than smart mammals. I big chunk of the determinant of intelligence is diet quality, which is why horses, cows koalas are so dumb and why koalas are so sleepy.

Doves, too. I’m convinced that doves lie, on the spectrum of intelligence, somewhere between bananas and sea slugs.

Living in a town with a sizable turkey population, they're also utterly oblivious to car traffic when they're in the middle of the road. It's interesting that they're smart about some things, and stupid about others.

I'm a developer, and I use computers in a way that gives normal people (including developers) the heebeejeebies. However my use of office stuff is relatively rare and extraordinarily basic - I would say pretty close to the majority of users' usage. There are no problems with libreoffice for that kind of usage. In some respects for people not hunting pointless features it's usability is better.

I have some legal and administrative issues at the moment that require me to share office docs fairly regularly, as well as editing PDFs for filling in forms. I have had no collaboration problems at all with either of these activities exclusively using libreoffice. At one point I tried out the online versions (google docs / office 365 online) and it was a disaster.


I'm not your demographic - committed amateur muso mostly acoustic (saxophone) with a side interest in electronic music. My seemingly sudden better understanding of the structure and uses of the diminished scale (have been looking into it on and off for a couple of years) may well coincide with the ease with which the Sloop firmware allows me to easily make interesting-to-me musical sequences. However most of my musical activities involves playing sax in medium sized or big bands. Mind you I can not understand song mode and how to save stuff yet, but that's a permanent problem I've had with every groovebox type thing I've tried.

Can you explain what exactly is the understanding that you gained?

it's kind of difficult to explain. Here's a typical playing by myself warm up to improvisation for me:

Start off by playing some arpeggio - say a major chord. Add the 7th - maybe test out chord V and VI to get a sense of the harmonic proximities (Coltraine's version of My Favourite Things does something like this). Modulate to the 4th, head back to the first and start off some bluesey musical language. At some point it gets boring - quite quickly - knocking out the pedestrian blues is easy - straight blues bands usually have some showiness / performative aspect to distract you from the boring bits. A bit of understanding of the diminished scale gives you more opportunities to modulate to different places in the harminic space and back because it guides you to the useful semitones and interesting dissonances. However it's usually fatal to think hard about this while you're doing it - in my experience it's more that something you've internalised in the past comes back as intuition.


it looks to me like it's authentication related, and I know from talking to people about vulnerabilities that don't become public that the 365 authentication systems can behave in quite "surprising" ways.

insufficiently supervised AI on top of poorly architected sales and marketing driven decisions looks like it's a good recipe for a nice warm garbage fire

I am aware that the authentication subsystems have major structural problems which make finding things like exfiltration bugs in azure surprisingly common. I suspect the current incident is related. I'm hoping it's also caused by too much insufficiently supervised AI coding on poor architectural foundations, and that being caused by that will be obvious and hopefully become public. But that's just speculation on my part.

I can't see a pathway for these things to be able to learn from experience in any meaningful way - the energy budget seems to prohibit it. Artificial neural networks are already many orders of magnitude more energy intensive than natural neural networks. Despite this, large nervous systems are very energy intensive as well. For example in humans 20% of the energy budget goes to the brain which is 5% of the body weight.

So I refrer to LLMs as language extrusion confabulation machines. Language extrusion was a term I heard the linguist Emily Bender use. Confabulation because my observation is that talking to an LLM is very much similar to my experience of interacting with Korsakov syndrome patients some years ago.

I look forward to the hype settling down to see what we end up with.


> Artificial neural networks are already many orders of magnitude more energy intensive than natural neural networks.

What metric are you using to compare? By most counts, the energy budget of an instance of an LLM in a datacenter is lower than the energy a person uses. Of course, if you count energy per neuron connections vs weights then you'll likely get a quite different number. But then again LLMs do a lot of things with far fewer weights than the brain does neuron connections, even if you only count neurons in some parts of the brain. And of course you can point to capabilities that the brain has but LLMs lack, but on the whole it feels like it's pretty difficult to make a useful like-for-like comparison here.


I can't make sense of your comment. Firstly because of the obvious massive over-build of GPU infrastructure the AI companies are engaging in. Secondly because the instance of the LLM in the data centre that users interact with is only a small part of the story. The training phase is clearly prohibitively expensive, thus the fact that these things have no way to learn from experience except by smoke and mirrors.

Also your comment feels like the classic climate denial discourse - say something a bit complicated and a bit difficult to follow that looks at a very small out of context part of the story to cast doubt.


Maybe I shouldn't have been pissed of by your commient. In which case - the excess costs of the training, output, feedback, and training cycle of artificial neural networks seem to prohibit the nightly training consolidation cycle (i.e. sleep) that is an important part of natural neural networks.


There's two sides to efficiency: one of them is how much you put in and the other is how much you get out. I was asking the question because the answer of the relative efficiency of a human vs an LLM strongly depends on what you put on either side of the equation. It's a really easy pitfall to look at one side of one version of that equation and presume a really high inefficiency when that's not necessarily the case. It doesn't help that the data necessary to fully answer any version of this isn't publicly available. That's why I was asking for something more concrete in terms of how you were making the comparison.

If you're looking at training costs, there is definitely one aspect in which LLMs are obviously significantly less efficient: the amount of information needed for the initial training. This does translate into some pretty high costs but it only needs to be paid once for the amount of work that any given LLM does. In terms of fine-tuning LLMs can get significantly more data-efficient than the initial model, which also means energy-efficient, and it's not obvious to me that it would be drastically worse than a human (though again, only thinking in terms of doing the energy input for the kind of work that an LLM is good at).

(The increased efficiency in comparison to humans is part of the reason why you see Jevon's paradox mentioned a lot whenever concerns about the resources used by LLMs are mentioned: more efficiency can easily mean more resource use in total)


Meta's bots are spectatularly awful, and if your organisation needs to use facebook advertising, they're also unblockable. There are ways around it that doesn't exactly block them but does clearly communicate they should GTFO, but it just shows you what an awful company Meta is.


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

Search: