Years ago I talked to an ex-msft guy whose comment on Samba was "they're repeating all of the same mistakes we made". It was a tongue-in-cheek comment about how compatible Samba was with Microsoft's SMB implementation.
Is this meaningful? Aren't frontier models MoE anyway? If you issue a StarCraft query they will probably just route to their StarCraft optimized model.
MoE is a misleading term, there is no "starcraft expert". There's not even a python expert. MoE just means a subset of blocks of weights are activated per query, nothing more. In so far as they might have legible functions they're like "balancing parens"
MoE doesn't work that way. It's not about increasing performance of a task, but speeding it up. To get a faster model, you need to make it smaller. If you make it smaller, the less it can remember. MoE tries to get faster responses without reducing the capabilities by running a smaller part of the model, and constantly swapping the parts used, so that the whole model gets to influence the output collectively, but each token is created by a smaller set of weights.
Anyone else feels like they can tell when the underlying model is MoE? Dense models are more concise in their answers. MoE seems to stall for time, writing vaugely until a right set of weights gets picked and commits to an answer. MoE also backtracks a whole lot more.
Elite overproduction I think refers to the unemployment rate of people who graduated from elite universities. I think that rate is probably still not high enough to be concerning for now.
My point is that any such metric may turn out to have been wildly off in a post-ASI world. Ten years ago or even today it may have seemed fine to educate a large part of the population for ”elite” jobs, but after ASI that may instead be looked at as a massive overproduction of ”elites”.
I have a few friends who used to really get into C&C games, and factional balance is basically nonexistent (with the resident 'good guy' faction being the best), and usually the winning strategy is to just mass produce whatever bang for buck unit that faction has (basic tanks most likely).
Sure, but looking at C's age, the language was changed significantly over time. With one of those changes, certain things could have been considered compile errors, not optional tooling.
Or if you turn your argument around: if people depend on workflows and automation (I do of course too), then this is a good reason to standardize the diagnostics and errors of the checking tools to the point where the workflows and automation can depend on them. Otherwise they will remain tools which are completely optional to run, or even worse, are not part of your normal work flow.
WireGuard is one of the easiest things to set up. The total amount of reading you'd need to do to set up WireGuard amounts to half of a page. I spent more time figuring out what firewall rules are needed for WireGuard traffic than actually setting up WireGuard, it's so simple.
I'm going to disagree with you there. It's been a while since I set up anything with WireGuard but "easy" is not a word I would apply to it. A simple tunnel or VPN configuration was close to that but a network comparable to a Tailnet got pretty involved to do properly. By contrast Tailscale was effortless.
Needing to do additional things like figure out firewall rules is exactly what I'm talking about. I enjoy tinkering for fun most of the time but I'm generally looking to tinker on something specific, not every step along the way.
Doesn't every display technology have issues with burn in? I have a VA monitor that has burn in as well, even though many have claimed that VA doesn't get burn in
I have a VA monitor that I bought in ~2019 with -according to its service menu- just under 40,000 hours of backlight-on time. It has some small hemispherical darkening along the top of its left edge that you can only notice when that portion of the screen is bright but has zero burn-in. [0] I use it for both video gaming and ordinary computing tasks.
[0] This darkening corresponds to nothing I've had on screen... it looks like the sort of minor backlight or diffuser (or whatever it's called) failure you tend to see on middle-aged monitors.
Same goes for drugs, tobacco, alcohol etc. oh look how that ended up. There's a reason why Prohibition has been repealed in every major country that tried it.
Alcohol is older than recorded civilization, while online gambling has existed for about half of one human lifetime, and has been mainstream for less than a decade. We're watching the negative consequences play out in real time. This isn't prohibiting something innate to humanity; you could still go to a casino and gamble before DraftKings existed.
10 years ago I had some friends who were into online sports betting. It required getting a VPN, accessing offshore servers, signing up for sketchy websites, and linking an online bank account. It was enough hoops to jump through that the average person wouldn't bother. Back then I would have agreed with you: let the people play! But I'm only 33 and I've watched online gambling cripple peoples' lives in just the last few years. It's a huge problem and should be made illegal again. You should not be able to access it in one click from the App Store.
Just like all prohibitions, if you take away access to X, people will just find other (often illegal/dangerous) ways to do X. We can make gambling less convenient and push it back into the backroom poker rooms and illegal bookies who break peoples' legs, but it would be more effective to solve the problems of despair, desperation and hopelessness that are driving people to gambling in the first place. That's a lot harder, so here we are talking about bans.
Gambling addiction has always existed but making it where you have to physically show up somewhere to do it is actually a highly effective deterrent vs. being able to click a button on your phone and get an app for it.
There is data from the past 10 years showing a direct spike in bankruptcies, divorces, foreclosures, suicide, and domestic violence (yes, really) as soon as DraftKings-style gambling is legalized in a state.
> despair, desperation and hopelessness
These are certainly factors, and yes we should have better mental health treatment and job opportunities in the US, etc.
But it's an incomplete picture of what's happening. There are billion-dollar companies with teams of advanced psychologists crafting the most addictive experience possible, and specifically targeting problem gamblers, with these new apps. It's industrialized and predatory. People with happy fulfilling lives and careers are still falling for it, not just hopeless impoverished people. Saying it's only about mental health and despair is framing it as an individual failing and ignoring the huge industry creating it this way.
The apps and the companies behind them really brought in a dangerous paradigm shift and the data is already here to show it.
> There are billion-dollar companies with teams of advanced psychologists crafting the most addictive experience possible … It's industrialized and predatory.
Sorry, I thought this was a discussion about gambling, not Facebook and TikTok and YouTube and Whatnot and OpenAI and EA and Roblox and every other company that invents a problem (oh no, your hair isn’t shiny!) and uses psychological manipulation to sell you a solution. :-)
Jokes aside, the fact that this sort of predatory behaviour is showing up in multiple industries with broadly similar characteristics (giant corporations with immense wealth concentrations who are not part of the communities they serve and whose only interest is in sucking up all the money) suggests that there are some deeper problems here, and that focusing on the gambling industry is focusing on a red herring (or perhaps the most obvious canary to date, to stick with bird metaphors).
> Saying it's only about mental health and despair is framing it as an individual failing
Not at all! Despair is the outcome of a broken system.
Strange response, do you believe that society has always had access to online gambling, in the same way that alcohol and drugs have been around forever?
So where was this multi billion dollar industry before? This demand is highly elastic and new, which you cannot say about any of the others you mentioned.
alcohol prohibition has always been difficult because it's so tied into food and social culture. Tobacco use has been extremely successfully restricted, harder drugs and gambling also remain prohibited in many countries.
You may not be aware of how rapidly and historically unprecedented gambling has spread in the US in the last few years. More than half of all men aged 18-49 participate in some form of online sports betting (https://futurism.com/future-society/men-aged-18-to-49-hooked...)
There are shades. Sweden allows the sale of alcohol, but it's tightly controlled. Still legal, but you have to buy it from government run stores, and they close early, and bars have to follow a bunch of laws too. Total prohibition doesn't work, yes, but that doesn't mean the government should also be totally hands off of it either.
I agree that Blindsight is highly overrated. Peter Watts devotes much of the book to ridiculous dialogue about the Chinese Room Experiment (astronauts in space apparently have nothing better to do than poorly discuss undergrad philosophy apparently).
The treatment of the Chinese Room Experiment is shallow and Watts clearly does not possess a more than rudimentary grasp of the philosophy, nor of computer science apparently.
It also felt shoehorned in. As was the vampire subplot, as other commentators have already discussed.
Felt like Watts just wanted to add in the Chinese Room Argument to make the novel seem "deep", when in fact it accomplished the exact opposite. No computer scientist that I know of takes the Chinese Room argument seriously, Scott Aaronson had a Quora answer where he mentioned how when he presented it to his undergrad class none of the students found it convincing, even after he tried to present arguments for it.
I don't like most of John Searle's work in general, but his Chinese Room Argument is obviously one of the worst pieces of philosophy of the 20th century, on par with the worst of the works ridiculed by Sokal et al.
As someone who literally studied consciousness and perceptual neuroscience in grad school, and ran studies on real-life blindsight patient GY, you couldn't be more mistaken about the Chinese Room's relevance.
The original formulation is awkward to modern ears, but the heart of the problem remains: How does/can consciousness emerge from smaller unconscious parts?
A single neuron can be modeled with a physics simulation, and is clearly not conscious. But at what size/configuration does a group of neurons become conscious?
"No computer scientist that I know of takes the Chinese Room argument seriously" Well, you now know one who spent years of his life studying consciousness.
Also, why would you even assume a priori that computer scientists have any worthwhile insights on the topic of consciousness? It's a completely different field. (I swear, the denigration of non-CS expertise on HN is maddening, as is the idea that every programmer is secretly a Renaissance polymath.)
---
What's funny is I read Blindsight in grad school, and also dislike it, but for very different reasons. A much better criticism is Watts's idea that human consciousness is useless or detrimental, because the aliens don't have it and don't need it.
First, the idea that consciousness confers no advantages is odd. The very word "blindsight" refers to doing a task without conscious awareness, but I can tell you from the zillion studies done on GY, blindsight task performance is nowhere near as good as sighted performance. Given Watts named his book after that, you'd think he might have known more about it.
Second, the book rests on the assumption that consciousness is cheap enough that its metabolic costs are irrelevant.
I don't recall if Watts states this explicitly, but I think the idea was that humans were the local maxima on earth due to other, non-consciousness advantages, and that most life in the galaxy didn't use it.
The problem is the brain is an incredibly expensive organ to maintain, consuming a huge fraction of our energy budget to maintain homeostasis. Forget the galaxy, if consciousness had no benefits (or was actively detrimental), then a sapiens variant without it on earth might easily out-compete the conscious variant and spread.
> As someone who literally studied consciousness and perceptual neuroscience in grad school, and ran studies on real-life blindsight patient GY, you couldn't be more mistaken about the Chinese Room's relevance.
> The original formulation is awkward to modern ears, but the heart of the problem remains: How does/can consciousness emerge from smaller unconscious parts?
> A single neuron can be modeled with a physics simulation, and is clearly not conscious. But at what size/configuration does a group of neurons become conscious?
This is clearly not the Chinese Room Argument. The Chinese Room Argument was supposed to prove that computers will never be capable of consciousness simply by following rules written down in a book. It was directed at symbolic AI systems that literally were just computers following rules in a book. Modern deep neural networks as well as human brains clearly don't function anything like that.
The idea that computers can't think is clearly based on irrational prejudice against electronic technology. Why can't a computer gain consciousness by simulating a human brain? John Searle himself said that consciousness requires "organic substrate" without any elaboration. Taken in context it's clear that the Chinese Room Argument was fundamentally clearly and obviously just motivated out of an irrational bias and prejudice against electronic/non-organic substrates.
Correct, but you're missing the forest for the trees.
As I'm demonstrating, the Chinese Room argument is still relevant regardless of Searle's original intent, because its underlying logic also applies to neurons and neural networks.
I restated it in modern neuronal terms because the exact same pieces/gestalt argument applies to biology.
Nobody still discusses the Chinese Room because they're interested in 80's symbolic AI. This is why I said the "original formulation is awkward to modern ears".
> Forget the galaxy, if consciousness had no benefits (or was actively detrimental), then a sapiens variant without it on earth might easily out-compete the conscious variant and spread
Isn't this pretty much what happens with the vampires? I know they're not completely non-conscious but they're clearly supposed to be somewhere between humans and Rorschach.
It's been 20 years, and I don't recall the details, so maybe?
I guess that would fit with Watts's idea of humans being a local maxima. IIRC, wasn't the vampires' dominance ended by the invention of architecture, given they had some sort of right-angle-based susceptibility to seizures?
Don't get me wrong, I enjoyed the book, even if there's parts I disagree with. I just think it fell short of being hard SF grounded in what we know already.
> IIRC, wasn't the vampires' dominance ended by the invention of architecture, given they had some sort of right-angle-based susceptibility to seizures?
Yeah, exactly. I think that was his way of answering "if human-level consciousness is such a waste of resources, why did humans become dominant on Earth".
I think if he had argued that consciousness was a "transitional power source for intelligent behavior" it would all be cleaner. It is and was useful, but eventually becomes a bottleneck.
That would have made an excellent addition to the book. I'd still be dubious, but it would have made for a more satisfying explanation.
That reminds me of a Philip K Dick short story along those lines, The Golden Man. It's about a good-looking mutant who has no verbal abilities, but can see all possible outcomes up to a point in the future. He breaks free of containment, uses his ability to seduce women (also to escape), and in the end, the govt officials tasked with capturing him lament that he might be the end of the human race, if he has children with his ability.
It's not about consciousness, but the story definitely supposes that evolution can jettison intelligence if a better ability came along.
We already spend as little consciousness as we can get away with.
Most of conciousness's effort is working itself out of a job, making plans to train the nonconcious bits of the brain to do jobs. Training and kicking off mental simulations.
In more intelligent folks, that learning and training often becomes a non-conscious behavioral pattern itself. I study and learn mostly non-consciously. I was trained how to.
I think most people spend most of the time as philosophical zombies and that isn't something to be scared of.
Very true, the brain is expensive! Unconscious processes are faster/cheaper, but usually less flexible. It's like we're constantly building FPGAs.
Most of human brain activity is unconscious, yes, but p-zombies don't quite fit us imo, since they have NO conscious awareness by definition. I really feel it's meant to be a more present/absent distinction.
When I was younger, I thought the Chinese room argument was ridiculously wrong. Any system that could do what it could do would have to be conscious.
Now with the rise of LLMs, I have to say that I was completely wrong. LLMs give us a simulation of consciousness, but they aren't conscious. My mistake was thinking that language was the hallmark of consciousness, which I no longer believe. It seems obvious to me that other mammals are conscious, even without language. Language is orthogonal to consciousness.
I still believe that there is nothing magical about biology that means that a computer program could never be conscious, though.
In my case, I found the Chinese room completely convincing when first exposed to it. Following an very complex code book to process a reply doesn't make the user conscious of the conversation.
Machines automate physical work without being conscious of doing it, and the way computers automate mental work, for example. a complex program, can't make them conscious. Truth be told, we don't even know what makes us conscious, we just know we are. We do know how computers work from the ground up, though, and it's all preset math.
>but they aren’t conscious
You don’t know that.
>language is orthogonal to consciousness
I don’t think that is true either. The beings we think of being conscious all are social creatures and communicate in some form.
Post on a board, deeply in a love/hate relationship with chinese rooms and regularly ascribing consciousness, emotions and personalitys to said chinese rooms.
reply