Matumio

The book makes a reference to Portia, which seem to be quite intelligent jumping spiders in reality, in the sense that they can plan long convoluted paths and may be able to count.

Research article: https://www.frontiersin.org/journals/psychology/articles/10....

You seem to talk about fairness. Copyright law isn't supposed to be fair, it's supposed to benefit society. On one side you have the interest of the public to make use of already created work. On the other side is the financial incentive to create such work in the first place.

So the question to ask is whether the artist would have created the work and published it, even knowing that it isn't an insurance to their family in case of their early death.

Yes it's very odd. Using non-real images is one thing (I mean we do that with 3D renderings), but images that don't show the product correctly is quite another. It destroys all trust I might have had for them as a seller, and all respect I might have had for them as a creator.

Interesting references, thanks. But both seem to be at the research and feasibility stage. From the ETH group's web page: (https://fml.ethz.ch/research/ses.html)

> To demonstrate the technical feasability of this process, we buildt a 10MWh pilot plant at ETH Hönggerberg. The first charing cycle, using hydrogen to reduce iron oxide to iron, was successfully completed over a time span of 4 months. The discharging cycle is currently ongoing.

So either they haven't managed to do a full cycle yet, or they are not updating their research page. It sounds like this should work, so I'm tentatively optimistic. But this looks like a technology you'd have to bet on, not yet a certain path to a commercial seasonal battery just waiting for mass deployment.

Production is trivial, but storage and transport really isn't: https://climate.mit.edu/ask-mit/can-we-use-pipelines-and-pow...

It's a very long stretch from "generate hydrogen" to "powering existing fossil fuel plants".

The most unlikely part is not even creating renewable fuels (that is a stretch already), but the idea that those fuels are going to be compatible with existing plants and infrastructure. It's not impossible, but it would probably be the least economical way to go about it. I recommend reading some industrydecarbonization.com articles for going a bit more in-depth about the why.

You mean everything without a big name attached. Being open source doesn't play into it. (Except that it gives you the theoretical option to verify and build it yourself, or to get it from a trusted third party who did.)

The chance to get 123456 in 100 actuations is very unlikely, indeed. Would you have had the same reaction for 654321 or 777777 or 121212? If there are a hundred such patterns, the chance of getting any in 100 actuations is 1%!

The sad thing is, there is no real obstacle to building a fast SPA. It's certainly not harder than to do it all "the old way". But you have to pay attention to performance.

If you just implement as many features as you can in as little time as possible, then the outcome will be something like Confluence, I guess, where you wait several seconds to load a documentation page you wanted to just read, not even edit.

It's worth to distinguish writing for yourself (to think) from communication.

Recently I discovered a lecture about technical writing in academics (from my "writing to think" folder: https://www.youtube.com/watch?v=vtIzMaLkCaM) that drove this point home: If you're an academic, your default mode of writing is "writing to think", and you have to make a conscious effort to avoid this mode when you communicate.

I think it also goes the other way: If you mostly write for others (or read what they write or say in public), then you need to make a conscious switch to "writing to think", which is a different skill where you worry less about being misunderstood and more about connecting to your personal knowledge.

For me the metaphor is DRAM and swap. To hold data in DRAM you have to pay the occasional refresh cycle, or else it may fade away. You swap your thoughts to paper (or disk) to get rid of that nagging feeling that you're about to forget something that could be important. This frees up brain cycles (and memory) to focus on the next thing.

Two days or years later you may skim your past thoughts, and there is some chance that it connects back to whatever is in your working memory right now. Then you can swap it all back in, and you have a chance to make a really non-trivial connection between related topics. Or you may notice that you are going in shallow circles around the same topic every few months, and now you can either put the topic to rest, or swap all instances back in and give them some serious thought and connection.

If you wonder whether you could at least capture CO2 directly from the combustion process (instead of out of thin air), well yes that's cheaper but still too expensive.

The current CCS projects use highly concentrated CO2 sources, while the usual combustion process will generate air with only a few percent of CO2. There was an article last year about the Hammerfest LNG plant. They have a CCS project nearby, but found it cheaper to electrify the plant: https://industrydecarbonization.com/news/is-carbon-capture-a...

If you read the mathematical theory of causality (e.g. Pearl), you'll learn that you must have the ability to make interventions "from outside" (at least in theory) before you can talk about causality. You have to define what is "inside" the system you study.

If you define everything to be "inside", then causality disappears because intervention disappears.

That's my impression too. There was a flock of academics and relatively non-tech folks joining a few years ago. Most didn't stay, and a few of those who did are now flocking to Bluesky.

But the others seem to be very much at home. They are not millions. They are probably prioritizing the small communities that have formed over maximum reach.

I would say when Facebook arrived. But it wasn't so much "shift to corporate-owned", it was more that it allowed non-techies to put stuff on the internet for the first time. Us techies, we already had our hand-coded html web pages hosted at some (probably commercial) provider.

I think the answer is "usability". Look at all the community-made, non-commercial projects. They tend to suck because they weren't built for you. They were built for people with similar high investment into the thing they do, for experts or power-users. For them it works.

So IMO the key question is how to find motivation or time or money to solve someone else's problem, without being forced to maximize the money-making part. Because by now we can see exactly what happens when money is the primary goal. Everyone starts with good intentions (solving a problem), but the incentives are so powerful. If you don't follow them you'll start to struggle, long-term, or get out-competed by someone who maximizes the money-making part of the job.

Kind of? Text search will also find commented out usage, docu, changelog, and my .org file with refactoring ideas from last month. (For better or worse.) It will also show when a class with the same method names exists elsewhere - which you want to know about if the naming is good. To find usage of a method called "update", LSP results will be better, but... that was kind of the point? That you don't name your methods "update" if they are important semantically.

And rg also works in projects with multiple programming languages (e.g. you're using a method from JS and want to find its Rust implementation in the webassembly module) and across project folders (instead of pointing you to a read-only copy of an installed module). But, if you spend 95% of your time in a single codebase with a single language... YMMV.

We can't agree whether Portia spiders are intelligent or just have very advanced instincts. How will we ever agree about what human intelligence is, or how to separate it from cultural knowledge? If that even makes sense.

Indeed. And the cute fox almost doesn't need the credits. When you recognize David Revoy's style on a project page, you know the project is probably a community-driven effort, and is worth checking out if you value that.

Can confirm. I've played around with differential evolution and a few others. Nothing was able to beat CMA-ES. At best, after some manual tuning it was just barely matching the untuned CMA-ES. (The task was to optimize a small neural network, gradients not easily available.)

CMA-ES is not suited for very large problems (more than a few hundred parameters), but there are several variants that can handle this, like sep-CMA-ES.

If by "modern" you mean a desktop/server CPU, the problem is the complexity that exists because of optimization for average throughput. Do you really know how long you will wait, worst-case, if three different cores flush the same cache line back to DRAM? Or maybe some niche hardware you use has some funky behaviour that stalls some hardware bus, or waits in the kernel for a millisecond, twice a month or whatever.

On the other hand, on FPGAs, deterministic timing is so very simple. Your output will not be a single clock cycle late even if something else goes wrong in logic running on the same FPGA (except through a connection that you control).

If you really want to know, OSADL has a QA Farm that monitors worst-case interrupt and scheduling latency for various CPUs and Linux versions.

I generally agree with you, if you want someone else to do the checkout for you because you don't like doing it.

But if you also buy clothes only made from hand-spun yarn, I mean, it was a problem for the yarn-spinsters back then, and by all means help them. But paying someone to do something that no longer needs doing is how you create bullshit jobs. No problem with spinning yarn for recreation, but if they have to do some complete nonsense to earn a roof and a meal... there are kinder ways to help.

Do you read your password manager's code every time it updates? Probably not, because you trust the author's reputation.

I wouldn't trust this page with my passwords either, but not because of the reasons that you mention. I haven't checked, but maybe it is simple enough to read the code in its entirety and then self-host? If so, nothing wrong with that.

Hm, it's probably true that recommendation algorithms do something similar already, training on "human likes" that were influenced by the previous generation. But "human language" is a richer medium to carry information.

I don't think you need to be independent or autonomous to develop a culture. And a lot of human culture was passed down over generations without understanding why it worked. We just imitate the behaviour and rituals from our most successful ancestors or role models.

If new LLMs can access the past generation's knowledge of how to please human evaluators, they will use it. It's not a deliberate decision by an "agent", it's just the best text source to copy from. This is a new feedback loop between generations of assistants, and it bypasses whatever the human designer had in mind. Phrases like "it is always best to ask an expert" will pop up just because you tuned the LLM to sound like a helpful assistant, and that's what helpful assistants sound like in the training data. You'd have to actively steer the new generation away from using their ancestral knowledge.

I guess it comes down to what your definition of "culture" is. There is no targeted teaching of the next generation, for example - but is this a requirement? I agree that talking about "machine culture" right now sounds like a stretch, but now I wonder what pieces are actually missing.

I think they meant culture in the sense of knowledge that gets passed down from one generation to the next. Not a human culture of using machines, but a machine culture of using human languages.

In theory. But if everything is in the password safe, the malware can just grab that and upload? And cover its traces. As opposed to patching every application/service you might use, and get access only when you use it.

Not everyone can afford to just walk away more than once or twice.

And people may perceive the uncertain alternative of not getting that job right now as much worse than it would actually turn out, and agree to stuff they don't really want. Like the point made in this short comedy scene: https://www.youtube.com/watch?v=-yUafzOXHPE

energy plus buildings are a thing: https://en.wikipedia.org/wiki/Energy-plus_building

The question is when you consider planet Earth "destroyed". Most likely it will remain blue and keep its atmosphere. Life will continue. It could be "destroyed" in the sense that humans sustainably sabotage their own long-term survival, or the survival of other species.

Short of a nuclear war, I don't think humanity will get close to extinction. But I think we are on a path to lose access to today's cultural knowledge (like microchips, vaccines, aviation). If the population is forced to shrink over the next couple of centuries, wars over fertile ground seem more likely than specialized global supply chains.

When I read "evolution strategy" I was pretty sure to find some variant of the Canonical Evolution Strategy (as in https://arxiv.org/pdf/1802.08842) or maybe CMA-ES or something related. But the implementation looks like a GA. Maybe the term means different things to different people...?

Then probably you know about NEAT (the genetic algorithm) by now. I'm not sure what has been tried in directly using combinatorical logic instead of NNs (do Hopfield networks count?), any references?

I've tried to learn simple look-up tables (like, 9 bits of input) using the Cross-Entropy method (CEM), this worked well. But it was a very small search space (way too large to just try all solutions, but still, a tiny model). I haven't seen the CEM used on larger problems. Though there is a cool paper about learning tetris using the cross-entropy method, using a bit of feature engineering.