Not necessarily. In my experience, getting the right person to prepare a company for an IPO or a sale is hard. Most buyers will do due diligence and besides 'slashing costs' and 'growing the company', there is a skill set for getting governance and compliance practices in place and as well as leading the roadshow for the sale which has some similarities to raising private capital. For instance, if you don't already have explicit policies for workplace safety and environmental practices (e.g. what do you recycle, water usage, etc), you will usually need to put these in place. (We invested in manufacturing and these were extremely important to us). If you are located in multiple jurisdictions, you need to be ready to demonstrate that you are in compliance with local regulations and pass the equivalent of "integration tests", prove you are in compliance across multiple jurisdictions where their rules may differ or seem to conflict. The CEO knows what needs to get done and has the rolodex to get the people to help the company get these things done for a sale because he has done this several times before and understands the things that can go wrong.
The problem with all these 'right to repair' advocates is that they assume that it is zero costs and that the manufacturers will eat that cost. No it will have a cost. And that is likely to disproportionally impact the cheapest phones and the poorest households.
> No it will have a cost. And that is likely to disproportionally impact the cheapest phones and the poorest households.
I'd actually argue the opposite.
Historically the poor have been the ones who repair their property rather than throwing it out and replacing it. It's far cheaper to repair than it is to replace.
Assuming the poor repair and the rich replace, as is tradition, then I would argue a modest price increase would represent the rich subsidizing the poor.
I'd argue this cost has always been there and was never zero to begin with.
Mass producing non-repairable, cheap throwaway items is an excellent way for industries to externalize a lot of the lifecycle cost of an item onto society.
So you get to choose at which point you'd like to pay for it. Directly and very visibly in an item's price (maybe even making you reconsider if a purchase is actually necessary). Or later, in taxes or a decrease in quality of life due to the increasing overexploitation of our resources.
> Mass producing non-repairable, cheap throwaway items is an excellent way for industries to externalize a lot of the lifecycle cost of an item onto society.
If you're against the externalization of costs, shouldn't you be advocating for measures that fully internalize such costs (eg. by charging a disposal fee at the point of sale), rather than merely trying to reduce it by making phones more repairable? That way consumers and manufacturers can decide for themselves whether the repairability is worth the extra cost or not.
To live in a world with repairable devices or in one without is a social consensus problem. We either are stuck in one game-theoretic equilibrium or the other. Game-theoretic equilibria cannot be defeated by "letting individual actors decide".
They are solved by social consensus, and in this particular case, by the passage of the right-to-repair law that forces the entire system from one equilibrium state to the other.
What's the "game-theoretic equilibrium" that's preventing repairable phones if consumers demand it? We have repairable versions of laptops and phones but they haven't really caught on outside of niche circles. Maybe people don't actually care about repairability?
There are two kinds of players in the market: a large consumer class, and a small collection of oligopolistic producers. The game-theoretic equilibrium is in the actions of the producers.
Customers have many different criteria by which they judge phones: price, performance, available apps, social desirability, repairability etc. I would say that repairability is of lower importance than other criteria.
Producers know that selling unrepairable phones produce more profits long term. Obviously repairable phones are also a few perfect more expensive. So from the 00s when phones were largely repairable, to today, producers have sequentially produced phones that are less repairable but a bit cheaper. Ad campaigns have been used to make thinner monolithic phones as more socially desirable [1]. This also changes how customers rate the importance of different criteria.
Any new producer who creates a repairable phone not only has to produce a more expensive phone, but also has to create a niche of customers, against the competitor ad campaigns, that value that repairability. Hence, that new producer is not profitable. This is the equilibrium which explains that lack of mainstream repairable phones in the market.
But of course, consumers want repairable phones, but only if everyone else gets them as well - so they aren't uncool. Which is why different consumer groups have advocated to lawmakers, and now we are getting a law that forces everyone to change in tandem.
[1] Not much different from the cigarette ads of the yesteryears.
>Any new producer who creates a repairable phone not only has to produce a more expensive phone, but also has to create a niche of customers, against the competitor ad campaigns, that value that repairability. Hence, that new producer is not profitable. This is the equilibrium which explains that lack of mainstream repairable phones in the market.
But if a repairable phone is cheaper for the consumer overall (the don't have to replace as often), why would this be a problem? Japanese cars outcompeted American cars because they were more reliable, and consumers recognized that. Shouldn't consumers jump at the chance of reducing their overall phone TCO by 50% or whatever?
Good question. (1) People's preferences are a function of time [1]. (2) And people find it difficult to estimate unexpected future costs.
So at buy time, people don't prefer repairability very high, which is some uncertain future cost. Which is why they make the short term decision to buy the cheaper sexier phone. Later on, when the phone starts to break down, they start caring about it a lot more. We know the latter is true because consumer group advocacy is determined by asking people what they want.
Another tactic phone manufacturers employ is that they don't talk about repairability anywhere, so most consumers don't even know phones are repairable. They think you just have to buy a new phone. It's like if your bowl broke, you would just go and buy a new one because you believe bowls cannot be repaired. But if you belonged in the right area/tradition of japan, you would just take the pieces of the bowl and join them together [2].
>So at buy time, people don't prefer repairability very high, which is some uncertain future cost. Which is why they make the short term decision to buy the cheaper sexier phone. Later on, when the phone starts to break down, they start caring about it a lot more.
You can make the same argument for cars, yet reliable japanese cars won out. Smartphones have been around for 15 years now? I think that's long enough for people to figure out how much repairs they need.
> We know the latter is true because consumer group advocacy is determined by asking people what they want.
And yet their purchasing decisions don't line up with what their survey replies are, by and large. It's not because of lack of choice, repairable phones exist, and have existed for a while now. Fairphone is on its 6th iteration now? Yet its uptake is lackluster. Between stated preferences and revealed preferences, I'm going with the latter.
>Another tactic phone manufacturers employ is that they don't talk about repairability anywhere, so most consumers don't even know phones are repairable. They think you just have to buy a new phone.
If you think consumers can't be bothered to do a "[phone name] repairability" search (despite the fact that they care about such a thing, as you claim above), and need to have the info spoon fed to them by the manufacturers, then maybe they don't really care about it?
> It's like if your bowl broke, you would just go and buy a new one because you believe bowls cannot be repaired. But if you belonged in the right area/tradition of japan, you would just take the pieces of the bowl and join them together [2].
Of course, in our modern economy, it makes little sense to fix broken bowls. They can be manufactured so cheaply and fixing it manually taxes so much time/materials that the juice isn't worth the squeeze. Something similar applies to phones, only that replacing a phone also comes with the additional benefit that it has a faster cpu + more memory + better camera. You don't get any of that by making phones repairable.
There are two "types" of right to repair. One is "you can't use the law to prevent me from fixing things", the other is "you have to make them easy to fix".
Based on the article, particularly the part that describes the bill [1], this is the former.
Amending the law so that it doesn't prevent repairing things adds absolutely no cost to manufacturing...
[1] "The bill would amend the Canadian copyright act, allowing individuals or independent repair shops to break digital locks in order to make software fixes."
Yeah, I think this is an important distinction. As much as I personally wish almost everything was designed to be repairable and that parts/instructions were easily available from the manufacturer, I'm actually pretty uncertain that those should be _requirements_. I really just want it to be legal to repair and modify anything. The manufacturer should be allowed to design any way they want (and I will continue to attempt to avoid manufacturers who make repair hard), but the most important thing is that if I _do_ repair it, that is recognized as fully within my rights and that the manufacturer can't use the law to punish me, up to and including modifying software, as long as it's for personal use.
Anything beyond that probably needs to be a signal from consumers that they _want_ (and perhaps be willing to pay a price premium for) repairable goods with available parts and documentation.
No, we all fully expect it to be expensive. Manufacturers didn't move away from repairable devices for nothing; it's cheaper to manufacture and design something that doesn't ever get taken apart. It also helps ensure that your customers switch to a new product once their old one breaks.
Considering that I've bought $15 Tracphones with a replaceable battery and SD card slot, I think those poorer households will be okay. It's better than the government buying them garbage that was designed to be obsolete from the start.
There's also a cost to making everything disposable and hard to repair. $300 to Apple to fix a button on an Apple Watch is one such cost, but there's also the waste created. Unfortunately those costs are harder to quantify and will take a long time to show up on a balance sheet somewhere.
A lot of us will be happy if a repairable phone costs me twice as much, but lasts 3 times as long because we won't have to throw the entire thing away and buy a new one when a tiny IC, a battery or the display fails. It will also reduce the damage to the environment which is already at the brink. These things are not harmless to recycle, even if they can be.
Add to this, the fact that many OEMs go for exclusive deals with the parts manufacturers where the parts are not allowed to enter mass market. This is to make repairs costly by creating artificial scarcity. This suggests that the true cost of repairable devices is not as exorbitant or harmful to the poor as these manufacturers and OEMs project it to be. It's more in the realm of disinformation.
>A lot of us will be happy if a repairable phone costs me twice as much, but lasts 3 times as long because we won't have to throw the entire thing away and buy a new one when a tiny IC, a battery or the display fails.
That was a generic statement - I wasn't implying that such devices don't exist. And yes, fairfone is a good example and we need to support brands like those. I'm currently in the process of gradually replacing old devices with repairable ones.
Another cautionary tale that too many of us ignore with our virtue signaling: "I support X because it sounds good for cause Y". Unfortunately, we often don't consider system or policy resilience - how will it hold up if people intentionally abuse its rules. Sometimes it doesn't help cause Y and has other negative effects.
I would put heritage building laws into a similar category of well-intentioned rules that can so easily be weaponized alongside overly strict environmental review.
In isolation keeping a neighbourhood character by setting rules around the paint colours and trim designs permitted seems like a benign set of laws to keep some interesting older neighbourhoods around. In practice they lock a city at a specific low density, often very close to the downtown core since the oldest development tends to be closest to the action. It can also exclude poorer residents (or even pretty well off people who can afford a $250,000 reno but not the $500,000 it'll take to satisfy the heritage committee).
I love an old victorian house, but not when there are hundreds of people living in tents next door and thousands more terrified they will have to join them because the cost of living is rapidly rising. If someone wants to pay to move that charming house to an area of lower average density, great. Otherwise it needs to come down to make way for hundreds of new units so people can actually afford the city.
The existence of heritage designations and the related restrictions also encourage developers to completely tear down any existing, older, undesignated buildings on their properties as soon as they acquire them, even if the buildings are still usable, and the new developments may be years away.
The developers don't want to risk the buildings being designated, or even just the designation process itself being used to introduce lengthy delays and extra costs, at some point in the future.
Rather than helping to preserve older buildings, the risk of a heritage designation just speeds up the destruction of them. In the end, there are fewer older buildings, and more empty lots.
Just like how the Endangered Species Act creates a perverse incentive not to report any Bald Eagle nests or even drive off any special animals. Also a rent control law that caps increases to CPI +2% results in landlords taking the maximum increase every year in response to the loss of optionality.
We really overuse these regulations. There are a relatively few buildings that are actually historic enough to warrant saving. But in many areas we're marking any old building as historic on the thinnest of pretenses. Having grown up in a city that was ruled by the historic review board, I'm unimpressed with the results. Maybe it's nice to see what they built 100+ years ago, but being forced to live in it forever? No. People change, neighborhoods evolve, we should carefully embrace that.
Sorry but in housing the value comes from "Location, ___location, ___location". If a neighbourhood can change into anything what kind of confidence do I have about the home that I'm buying? Things are already variable enough as is. Maybe if zoning changes had a 20 year window from being enacted to being effected. That might balance out needs.
(In case it's suggested otherwise, I live in urban high-density housing, I'm just sympathetic to the concerns of others)
The answer is quite clearly that it is immoral to allow those who can afford homes from being protectionist about their assets when it comes at the expense of those who lack affordable housing.
The need for housing outweighs the desire to get a return on your investment.
And who is the arbiter of this morality or the urgency of this need? I say it would be elections where everyone in that area gets a chance to vote and decide for themselves.
And as we're seeing here, if they can't get what they want one way they do it another way.
I've always thought the way out (besides just waiting until a neighborhood is an absolute shithole of a slum and can be redeveloped because poor people have little political power) is just to straight up bribe people. New developments have a "fee" that is directly applied in cash to other homes in the area to reduce property tax.
I think in most US states, the legal situation is actually that local governments are creations of the state. They are allowed to have their own ordinances and so on as a matter of convenience, to avoid state legislators having to bother worrying about every edge case that only comes up in one county, but they do not have a right to exist independent of the state saying that they do. There are exceptions, but municipal ordinances (and HOA rules) can be overridden by state law.
We don't /have/ to let all these little NIMBY fiefdoms exist. They exist at the pleasure of the state legislature, and therefore voters statewide, not just locally (modulo gerrymandering, a big caveat).
Study after study shows that the value of property goes up when density increases, not the other way around. People who cling to property values as a gatekeeper are actually arguing against their self-interest, usually without knowing it.
> People who cling to property values as a gatekeeper are actually arguing against their self-interest, usually without knowing it.
That is rarely the reason, in my experience. It is merely an accusation that gets thrown around by people who want the NIMBYs to look shallow. There are usually far more specific reasons for opposing new development.
There are always just a few, however, who do want the money. So they cash in and sell out to developers, and then move to a less dense neighborhood. Eventually the whole neighborhood character does in fact change, it just takes years.
That's going to make the food that the poor person is ultimately buying from the billion dollar supermarket pretty expensive...
Supermarkets are, generally, ground zero for problems with corporate taxes on things other than profits. They're extremely low-margin, so proposed revenue and transaction taxes would generally just increase prices of everything in the supermarket by whatever the tax is.
(That's just one problem, of course, but it's a very obvious one.)
Well, profits or VAT. Even more if you make some actually sane VAT that discounts all VAT paying expenses and all labor expenses.
I'm not sure any VAT on the world fits that description, but it is the one way to make them work as well as profit taxes, but as easy to tax as any other VAT.
Ahhh. Good old Fourier analysis on the change in model parameters: "low-pass filters that screen out background noise, high-pass filters that help analyze background signals, and Gabor filters that are often used in image processing. The Fourier analysis of the kernels revealed the neural network’s parameters were behaving like a combination of low-pass, high-pass, and Gabor filters."
I can't imagine the architects of the model didn't design the layers and kernels without some idea in mind of what sorts of operations they would be wanting to simulate right? Do people just throw random layers and activation functions together until something works? I'm only on the periphery of dl but those I know that work in it either have a general idea of what equuations they would want to use to do it without dl, or have a biological inspiration in mind. This doesn't seem like a biologically inspired sort of ___domain.
> Do people just throw random layers and activation functions together until something works?
In a lot of cases, yes. You can start with a reasonable baseline architectural guess, like “convolution should be good for vision” or “attention should be good for language”, but after that it's a lot of guess and check.
This may be the case in certain shops or disciplines, but I can assure you, if you have the right insights, you can use statistical and ML principles to achieve better accuracy with orders of magnitude less compute.
I consider it a tragedy that we throw huge hogs of models into datacenters and let them churn without giving much thought to improving performance. The climate impacts relative to the real benefits are measurable and depressing for a field so focused on innovation. Usually the 100x model built by a team of "data scientists" over a month is no better than a 1x model built by a couple SMEs over a couple weeks.
Just because there are a lot of under-qualified bootcamp data scientists out there who think Pytorch can solve all their problems, that doesn't give you the right to sneer at the cutting edge of ML research on the grounds that you've built some good models in Excel in the past.
There are plenty of times when deep learning is the wrong solution, and probably most businesses out there should not ever expect to need any kind of highly compute-intensive model training. But there are also plenty cases where neural networks are the right solution as well (even if not "deep learning" as such).
Maybe I'll be convinced if you share some specific examples. Otherwise I'll assume that your comment is just arrogant derision.
I have a lot to say on this and related topics, but I'll try to keep it on topic to this thread.
> sneer at the cutting edge of ML research
I was reacting specifically to mis-specifications of what "good" means in these contexts, e.g., "attention should be good for language" from GGP comment.
The major advances of these models du jour are not the outputs, but rather the infrastructure that it takes to run them. I salute all those involved in implementation. I hope (and see!) these innovations spill into things that help people and generate real value, for instance, all the things businesses use accurate climate forecasts for.
My issue mostly revolves around the discrepancy between value-created-per-FLOPS and cost-per-FLOPS. If we run huge models for what serves for us the same purpose as the TV walls in Fahrenheit 451 do for their owners, what are we adding? If we are constantly running pixel-level segmentation networks on the cameras of automobiles that are already being driven by humans, how is that helping?
Stable Diffusion is a real innovation in optimization techniques. This is the only part of recent news that can be considered "cutting edge ML research", as it is an important step in unifying optimization theory with that of non-equilibrium statistical mechanics. The parallels have been noticed there for decades, and the cutting edge of ML research today is unifying them by cribbing aspects of each into the other.
The ideas (models) themselves are quite naive. The perceptual space of these models is essentially one-dimensional. This, in the broad strokes, is a terrible model for reality, and so these casual discussions of "intelligence" seem mis-informed, not to mention under-determined. It seems like a lot of mis-informed people speak quite confidently about the things they don't know that they don't know, especially on a site where people who value their autodidactive skills tend to congregate.
Inductive bias is the primary term of art that is missing from these discussions. The models du jour do not have any architectural components which represent language per se, only flat representations thereof. The reason LLMs work so well is because of the Transformers inside; multi-head attention is a clever trick to vastly expand the possible data processing techniques as information passes and is distilled through the network into the outputs, but ultimately it is just a fancy way of doing a technique we could call "quasi-random projections" under the ___domain of linear algebra.
The notion of intelligence we entertain in these discussions is hopelessly anthropocentric, and, furthermore, incredibly arrogant. The recent achievements say much more about how easy it is to simulate human-comprehensible language and images than it does about how we are birthing new intelligences. I believe the hype and deployment of these tools is doing more to hurt society in the long term than it is to help it, and I welcome more candid discussions about how simple the rules that make us "intelligent" might be. That seems to me, during my studies, to be the real question.
It kind of sounds like you're saying that it should be possible to build a better GPT-3 by incorporating more linguistics ___domain knowledge into the model somehow, beyond stacking word embeddings and transformers together. If so... then okay, show me.
The biggest innovation of deep learning is that you can take a relatively general architecture and build surprisingly sophisticated models by training that general architecture on a ton of data, that absolutely wipe the floor with the older style of models.
AlphaGo/Zero is maybe an exception here: it's not "just" a model, it's a whole system based around a model, which is specifically designed to perform well on certain tasks, and succeeds where a more naive system would probably fail. You could argue that ChatGPT itself is an early kind of advancement along those lines, going beyond just cranking predictions from a model to a more holistic task-specific system. But it's still mostly "just" a huge model trained on a huge pile of data, and I'm not going to argue strongly that it isn't. What I am arguing is that people aren't as naive as you think, and it's not for lack of trying that LLMs don't incorporate detailed knowledge about linguistics, psychology, et al.
You suggest that transformers shouldn't be considered "good" for text, because they don't include enough inductive bias. Don't they? They're really clever things in my opinion, and I think they actually represent quite a lot of inductive bias compared to what came before, and have a lot of intuitive appeal in their respective intended task domains. I hardly think it's fair or correct to consider the transformer nothing more than a linear algebra trick.
I suppose you're arguing that they aren't "good" solely because they produce state-of-the-art results and beat literally everything else we have. I think your argument is that they only work as well as they do because they need a tremendous amount of computing power and data to get useful results out of them. That might be true, but that's literally why people are trying to make them faster and less costly to train!
Don't forget that the other big innovation of deep learning is transfer learning and fine tuning. A big LLM like GPT-3 only needs to be trained fully every once in a while, while it can be reused and adapted to a huge variety of tasks relatively efficiently and cheaply. The whole point of the article at the top of this thread is making big models cheaper and faster to train. If you're bothered by the cost- and compute-efficiency of these things, what could be better than making the training process more efficient?
Moreover, it's kind of weird to insinuate that the field doesn't care about making these things faster or cheaper to run, when immediately after any new model is announced, a flood of projects follows trying to shrink the model, improve inference speed, etc.
Finally, who said anything about intelligence? I didn't, and nobody else did either. That's a completely unrelated topic as far as I'm concerned, and I leave that one to the philosophers to debate.
Convolutions are learned by the neural network. In general, how layers are composed is more of an educated trial and error process than of an intentional design producing expected results.
Yes, but the other 85% of the sectors would likely benefit a reduction in costs (i.e. insurance premiums for their employees) that would positively affect their bottom line.
