I'll be the first to admit this isn't my area of expertise. But after following these developments peripherally, I've been holding off buying a new desktop for awhile.
It seems like Intel has bumbled this at every step. They've put out a lot of misinformation causing a lot consumer confusion. It seems like every time they exclaim "it's fixed!" researchers say that's not the case. I'm assuming at this point we're probably at least a couple of CPU generations away from Intel fixing this properly.
On top of that, they've also been fighting the 10nm battle. More empty promises and missed deadlines on that front as well.
When I compare my current aging Intel system to single thread performance of the latest generation, it just doesn't justify the cost. AMD claims Zen 2 will fix all their problems. If they deliver, I will probably switch back to AMD. Intel burned a lot of goodwill in the past few years.
It's funny how the tech world works. Most tech companies thought html5 + javascript was the future. It makes complete sense from a rational point of view. Then the developers complained they needed native and then slowly they all caved.
Now I think developers are finally realizing that their app to display octopus recipes doesn't need the same performance as a first person shooter. ES6 and ES8 have made javascript tolerable, and typescript is now mainstream. There's been a battle for the fastest javascript engine and javascript far outpaces other scripting languages in terms of execution speed.
It always seemed bonkers to me that I would have to download an app running java or objective-c to display essentially a webpage. We would never tolerate this on the desktop, but somehow it became the norm on mobile. I'm glad PWA is making progress (though apple seems to want to sabotage it). Realistically, looking at my phone right now, I have maybe 2 apps that actually need native. The rest are glorified web pages.
Yes. The more I read about large tech corporations entering the blockchain space the more I realize they just don't get it. Their solution is to centralize and add authentication and authorization. At that point, why even use blockchain?
For anyone actually paying attention, we are in the most interesting times in blockchain's short history. We are starting to see layer 2 and sharding mechanisms speed up the networks by several orders of magnitude. Zero knowledge proofs are allowing smart contracts to interact with encrypted data. It's a very exciting time. Most of the interesting research is being done by startups and international researchers. It feels like the web in about 1995.
And large corporations just don't get it. The whole point of blockchain is to cut out the large tech middlemen. A blockchain network with centralized middlemen is just a worse system than what it attempts to replace.
You are completely correct. I can't get in the head of Satoshi Nakamoto, but it seems like he envisioned bitcoin as a currency and many modern bitcoin owners see it as an investment. Bitcoin network changes are slow and there just doesn't seem to be a lot of interest in scaling to millions of transactions per second.
OTOH, there are several cryptocurrencies that aim to be less volatile and some can scale to a million transactions per second (or at least claim to). Who knows which will succeed, but that seems way more interesting to me. I'd rather invest in those companies the same way one might have invested in Red Hat for open source software.
Ethereum is interesting to me because distributed apps are possible. The current implementation is incredibly wasteful. However, the Ethereum community seems way more interested in scaling than the bitcoin community.
I do a few each year as well and have found them to be very helpful.
That being said, the quality varies considerably. Coursera has constant issues with auto graders and the programming assignments often consist of inserting a few lines of code into a 200 line program. I've taken courses where the assignments have almost nothing to do with the lecture. Most new courses I've taken have mostly abandoned peer review because it has its own laundry list of issues.
This has basically been my experience as well. One of the big problems I've seen is there's a huge initial excitement for neural networks, but as soon as I tell people how much data and computing power is required to actually create a half-way decent model, it usually puts a damper on their excitement. This is somewhat changing with pre-trained and partially-trained models. However, most pre-trained models seem to be for image data.
The biggest issue I've had with spreadsheets are unnamed cells. G3+F1+C2 etc etc gets out of hand way too quick. Yes, you can name your cell variables, but very few people do.
A more effective combo imo is python, pandas, and jupyter notebooks. Python is a simple language, pandas data structures lend themselves to the types of calculations you'd do with a spreadsheet, and jupyter notebooks allow one to tell a sequential story as to how a calculation was done.
Just to be clear, the system that beat Ken Jennings has very little to do with Watson in its current incarnation. Much of the team was splitup after the Jeopardy demonstration and IBM decided there wasn't much market for a question/answer system such as this.
So they reused the name, banking on the general population having heard of Watson Jeopardy to drive sales. Watson in its current incarnation is actually mostly off-the-shelf open source and existing IBM tools. Apache Spark and IBM's SPSS are currently under the Watson umbrella.
Watson Jeopardy was interesting for the time, but other companies are doing much more interesting things these days.
I always look at bad reviews first. If the complaints seem legitimate, I decide if it's a deal breaker. There seem to be far fewer fake bad reviews than good. It makes it a lot easier to spot things like your Dell laptop charger scenario.
Regardless, my online shopping has slowed to a trickle. It's too much of a pain to return when it's garbage.
I haven't done Windows administration in a few years, but I'm on the receiving end of Windows update on two PCs as a user. Apt is a brain dead simple package manager compared to the sophistication of windows update.
In practice, apt is much faster and more predictable. I just received notice of updates for about 60 MB of updates in ubuntu. These took less than a minute to download and install. During the process, I was treated like an adult and the package manager let me know which package it was working on and whether it was downloading or installing. One of the most frustrating things with windows update is not knowing how long it will take. The second most frustrating thing is not knowing if it's stuck or just working on something that takes a long time.
To be fair, there have been many developments in the past few years that make neural networks considerably more practical to use. There's a lot of hype and marketing, but in some domains it's deserved.
(1) On the hardware side, it turns out many of the advances in GPUs are also really useful for training neural networks. The typical speedup on even a low-end GPU is at least 10x. This has spurred research into ASICs like Google's TPU which has yielded even bigger gains.
(2) The amount of large labelled data sets like ImageNet has exploded. Neural networks outperform many other algorithms when fed huge amounts of data.
(3) There have been some algorithmic developments which handle long-standing problems such as vanishing/exploding gradients, local minima shapes in hyperplanes, and neural network architectures.
(4) Libraries like Tensorflow and Pytorch have made them much more accessible to the average programmer. With Keras (built on top of tensorflow), the network is essentially legos that you piece together.
I think we have to be a little more formal with terminology. The summary and most articles these days use "algorithm" and "AI" interchangeably. You can use an algorithm to train a machine learning model, but the model isn't really an algorithm in the classical sense.
The trained model can definitely have bias based on the training data. The classical example is, train a word2vec or glove model on the texts of wikipedia, then find the vector representations of doctor and nurse. You'll find that nurse is considered a female term while doctor is male.
This may be acceptable for trivial things like advertising or movie suggestions, but machine learning is now being used for important things like job application screenings. Many times the model can be very opaque and this bias may not seem obvious. Even worse, it seems every company now wants to have AI in their product, and may have half-rate data scientists that graduated from a data science bootcamp.
The research I've seen on this subject is serious work. In the case of the doctor/nurse vector representation, the goal would be to make the occupation gender neutral. The tricky part is that you'd still want the model to retain certain qualities, like mother being female and father being male.
This update took it upon itself to create a new recovery partition, which then complains it's full. I eventually fixed it, but it seems like there wasn't very much QA involved in this update.
I think the hysteria is generally rooted in modern journalism being for the clicks and reality just gets in the way of that. If you believed the media, skynet is just around the corner. When you talk to researchers, they're working on boring things like vanishing gradients. There's a huge disconnect.
I think the world would be a better place if the terms "AI" and "Neural Networks" were never coined.
Much of this terminology is rooted in research from 50 years ago when we thought we understood the brain. People thought we could mimic this using math and code. It turns out, we understand very little of the brain and while neural networks do work incredibly well, they probably only work superficially like the human brain.
I somewhat work in this field and know a fair amount about machine learning and AI. I don't know anyone in the field that is worried about any of this (they may exist, but I can't remember meeting any).
I think the combination of a terrible name (AI implies this technology can "think" and "understand") and imaginations have really added fuel to the fire. There are two main reasons why this is not really a worry.
(1) The AI we have today is extremely primitive. When most people talk about AI, they are talking about the improvements in neural networks that have happened in the past few years. Neural networks can find optimal statistical associations. They are generally just a logistic regression, many times, stacked vertically and horizontally. Fancier networks generally just reconfigure the network architecture (i.e. RNNs) but they all basically work on the same principle.
They generally work by taking a set of inputs (images, audio, sensor data) and known outputs, then they optimize a set of weights that can best predict that output. That's it. This idea that they will somehow decide it's in their best interest to kill humans is far fetched. I'm much more concerned about a rogue dictator deciding it's in his best interest to nuke us. Andrew Ng once made a comment that we should be about as worried of self-aware AI as we should be about over-crowding on mars.
(2) There are many machine learning tasks, but most generally spit out an answer that has to be carried out by plain old boring code. For example, a model might spit out the best ad to show a user, but code has to fetch and display that ad. Even if AI somehow magically becomes self-aware, it's limited by the code we write. So it may magically output "kill all humans", but the only valid choices are "shoe ads" or "purse ads."
In the case of military robots being controlled by AI, it would be trivial to create out-of-band kill switches.
The bigger worry with AI is job displacement. Although I do question whether many companies can get their act together enough to make products that will actually displace jobs. Sure, google can make incredible virtual assistants using some of the best talent in the world. The job displacement will occur in markets much too small for Google to care about. So it will be up to smaller companies to write the software. In my experience, most of these companies can barely get a basic CRUD web app working, let alone a complex neural network with many highly tuned hyper-parameters.
I know a lot of people have ideological objections to WSL, but from a practical standpoint WSL isn't even very good. I tried it for about two weeks before abandoning it.
First, windows has a terrible terminal emulator. I don't think it's improved since Windows 95. Basic stuff like copy/paste is not intuitive, let alone nice features like tabs. I tried an alternative (cmder I think) and it was OK, but something as important as the terminal emulator should not be an afterthought.
Raw sockets didn't seem to work correctly (or at all). I tried a few network tools and they generally fell flat on their face.
It seems really slow. Maybe it's just my imagination, but sometimes I'd do something as simple as an 'ls' and patiently wait.
There was no GUI support out of the box. I had to setup Xming on the windows side. Again, not super complicated, but it seems like little thought was put into it. I don't need a GUI very often (usually just to display plots I generated), but there should have been more effort.
The goal was to basically have python, R, a C compiler, some networking tools, etc, available when I am in Windows and not have to boot a Linux box for basic things. The quality was just too low and went back to using a combination of VMWare and native windows versions.
Maybe it will get better, but it seems like it's trying to solve a problem most people don't have.
One of the things that has irritated me on occasion is their use of topic algorithms. I'm fairly certain that searches actually use an algorithm like Latent Dirichlet Allocation when you search. So basically, if I search for bad reviews of Dell the algorithm may choose to accept a word belonging to the same topic (such as negative instead of bad). If you've ever seen your google search bold a term you didn't actually search, that would be LDA at work. Most of the time it's useful.
This is fine for trivial searches, but it gets very irritating when I'm searching for very specific words. Is there a way around it? Maybe. But the hide-all-complexity user interface of google tends to discourage finding it and I just use another search engine.
There was a point in time for which having a membership made sense. This year is going to be my last year of prime. I've actually stopped using Amazon in general for most things in the past year.
Amazon relies heavily on third party sellers these days. It seems like the products from these sellers that are prime eligible just have the shipping costs built into the price. For example, I searched on Amazon "adafruit" (a hobbyist electronics company) and one of the first results is the "Adafruit 328 battery." $18.35 on Amazon, $14.95 on adafruit.com.
For certain items, prime (and Amazon in general) has good deals. If I'm going to by a popular $500 electronic product, Amazon w/prime is usually your best bet. But there are huge categories of products for which Amazon no longer (or never did) makes sense.
I remember an NPR story awhile back that many 3rd party sellers were simply buying products on ebay, marking up the price, re-selling on Amazon, and making a killing. People have been trained to use Amazon even when a simple search can find products cheaper from other sources.
Pantry sucks. The few times I've used it half the cans were severely dented and the deals weren't anything great. Basically Wal-Mart prices.
The prime streaming service is just duplicates of content Netflix and Hulu already offer. I'm guessing they just get the cheapest content (like 90's sitcoms).
Ironically, I've gone back to being a fuddy-duddy and just buying my products in-store. I get what I need and leave. It seems like I have far fewer impulse buys in-person as well.
This is not 100% true. There's a range of how interpretable models are. Decision trees tend to be the easiest to interpret, then linear models, and all the way at the other end of the spectrum are DNNs. But there's been a lot of research lately regarding making DNNs more interpretable as well.
This is not 100% true. There's a range of how interpretable models are. Decision trees tend to be the easiest, then linear models, and all the way at the other end of the spectrum are DNNs. But there's been a lot of research lately regarding making DNNs more interpretable.
Slashdot Mirror
I'll be the first to admit this isn't my area of expertise. But after following these developments peripherally, I've been holding off buying a new desktop for awhile.
It seems like Intel has bumbled this at every step. They've put out a lot of misinformation causing a lot consumer confusion. It seems like every time they exclaim "it's fixed!" researchers say that's not the case. I'm assuming at this point we're probably at least a couple of CPU generations away from Intel fixing this properly.
On top of that, they've also been fighting the 10nm battle. More empty promises and missed deadlines on that front as well.
When I compare my current aging Intel system to single thread performance of the latest generation, it just doesn't justify the cost. AMD claims Zen 2 will fix all their problems. If they deliver, I will probably switch back to AMD. Intel burned a lot of goodwill in the past few years.
It's funny how the tech world works. Most tech companies thought html5 + javascript was the future. It makes complete sense from a rational point of view. Then the developers complained they needed native and then slowly they all caved.
Now I think developers are finally realizing that their app to display octopus recipes doesn't need the same performance as a first person shooter. ES6 and ES8 have made javascript tolerable, and typescript is now mainstream. There's been a battle for the fastest javascript engine and javascript far outpaces other scripting languages in terms of execution speed.
It always seemed bonkers to me that I would have to download an app running java or objective-c to display essentially a webpage. We would never tolerate this on the desktop, but somehow it became the norm on mobile. I'm glad PWA is making progress (though apple seems to want to sabotage it). Realistically, looking at my phone right now, I have maybe 2 apps that actually need native. The rest are glorified web pages.
Yes. The more I read about large tech corporations entering the blockchain space the more I realize they just don't get it. Their solution is to centralize and add authentication and authorization. At that point, why even use blockchain?
For anyone actually paying attention, we are in the most interesting times in blockchain's short history. We are starting to see layer 2 and sharding mechanisms speed up the networks by several orders of magnitude. Zero knowledge proofs are allowing smart contracts to interact with encrypted data. It's a very exciting time. Most of the interesting research is being done by startups and international researchers. It feels like the web in about 1995.
And large corporations just don't get it. The whole point of blockchain is to cut out the large tech middlemen. A blockchain network with centralized middlemen is just a worse system than what it attempts to replace.
You are completely correct. I can't get in the head of Satoshi Nakamoto, but it seems like he envisioned bitcoin as a currency and many modern bitcoin owners see it as an investment. Bitcoin network changes are slow and there just doesn't seem to be a lot of interest in scaling to millions of transactions per second.
OTOH, there are several cryptocurrencies that aim to be less volatile and some can scale to a million transactions per second (or at least claim to). Who knows which will succeed, but that seems way more interesting to me. I'd rather invest in those companies the same way one might have invested in Red Hat for open source software.
Ethereum is interesting to me because distributed apps are possible. The current implementation is incredibly wasteful. However, the Ethereum community seems way more interested in scaling than the bitcoin community.
I do a few each year as well and have found them to be very helpful.
That being said, the quality varies considerably. Coursera has constant issues with auto graders and the programming assignments often consist of inserting a few lines of code into a 200 line program. I've taken courses where the assignments have almost nothing to do with the lecture. Most new courses I've taken have mostly abandoned peer review because it has its own laundry list of issues.
This has basically been my experience as well. One of the big problems I've seen is there's a huge initial excitement for neural networks, but as soon as I tell people how much data and computing power is required to actually create a half-way decent model, it usually puts a damper on their excitement. This is somewhat changing with pre-trained and partially-trained models. However, most pre-trained models seem to be for image data.
The biggest issue I've had with spreadsheets are unnamed cells. G3+F1+C2 etc etc gets out of hand way too quick. Yes, you can name your cell variables, but very few people do.
A more effective combo imo is python, pandas, and jupyter notebooks. Python is a simple language, pandas data structures lend themselves to the types of calculations you'd do with a spreadsheet, and jupyter notebooks allow one to tell a sequential story as to how a calculation was done.
Just to be clear, the system that beat Ken Jennings has very little to do with Watson in its current incarnation. Much of the team was splitup after the Jeopardy demonstration and IBM decided there wasn't much market for a question/answer system such as this.
So they reused the name, banking on the general population having heard of Watson Jeopardy to drive sales. Watson in its current incarnation is actually mostly off-the-shelf open source and existing IBM tools. Apache Spark and IBM's SPSS are currently under the Watson umbrella.
Watson Jeopardy was interesting for the time, but other companies are doing much more interesting things these days.
I always look at bad reviews first. If the complaints seem legitimate, I decide if it's a deal breaker. There seem to be far fewer fake bad reviews than good. It makes it a lot easier to spot things like your Dell laptop charger scenario.
Regardless, my online shopping has slowed to a trickle. It's too much of a pain to return when it's garbage.
I haven't done Windows administration in a few years, but I'm on the receiving end of Windows update on two PCs as a user. Apt is a brain dead simple package manager compared to the sophistication of windows update.
In practice, apt is much faster and more predictable. I just received notice of updates for about 60 MB of updates in ubuntu. These took less than a minute to download and install. During the process, I was treated like an adult and the package manager let me know which package it was working on and whether it was downloading or installing. One of the most frustrating things with windows update is not knowing how long it will take. The second most frustrating thing is not knowing if it's stuck or just working on something that takes a long time.
To be fair, there have been many developments in the past few years that make neural networks considerably more practical to use. There's a lot of hype and marketing, but in some domains it's deserved.
(1) On the hardware side, it turns out many of the advances in GPUs are also really useful for training neural networks. The typical speedup on even a low-end GPU is at least 10x. This has spurred research into ASICs like Google's TPU which has yielded even bigger gains.
(2) The amount of large labelled data sets like ImageNet has exploded. Neural networks outperform many other algorithms when fed huge amounts of data.
(3) There have been some algorithmic developments which handle long-standing problems such as vanishing/exploding gradients, local minima shapes in hyperplanes, and neural network architectures.
(4) Libraries like Tensorflow and Pytorch have made them much more accessible to the average programmer. With Keras (built on top of tensorflow), the network is essentially legos that you piece together.
I think we have to be a little more formal with terminology. The summary and most articles these days use "algorithm" and "AI" interchangeably. You can use an algorithm to train a machine learning model, but the model isn't really an algorithm in the classical sense.
The trained model can definitely have bias based on the training data. The classical example is, train a word2vec or glove model on the texts of wikipedia, then find the vector representations of doctor and nurse. You'll find that nurse is considered a female term while doctor is male.
This may be acceptable for trivial things like advertising or movie suggestions, but machine learning is now being used for important things like job application screenings. Many times the model can be very opaque and this bias may not seem obvious. Even worse, it seems every company now wants to have AI in their product, and may have half-rate data scientists that graduated from a data science bootcamp.
The research I've seen on this subject is serious work. In the case of the doctor/nurse vector representation, the goal would be to make the occupation gender neutral. The tricky part is that you'd still want the model to retain certain qualities, like mother being female and father being male.
This update took it upon itself to create a new recovery partition, which then complains it's full. I eventually fixed it, but it seems like there wasn't very much QA involved in this update.
https://borncity.com/win/2018/05/02/windows-10-v1803-update-creates-a-new-oem-partition/
I think the hysteria is generally rooted in modern journalism being for the clicks and reality just gets in the way of that. If you believed the media, skynet is just around the corner. When you talk to researchers, they're working on boring things like vanishing gradients. There's a huge disconnect.
I think the world would be a better place if the terms "AI" and "Neural Networks" were never coined.
Much of this terminology is rooted in research from 50 years ago when we thought we understood the brain. People thought we could mimic this using math and code. It turns out, we understand very little of the brain and while neural networks do work incredibly well, they probably only work superficially like the human brain.
I somewhat work in this field and know a fair amount about machine learning and AI. I don't know anyone in the field that is worried about any of this (they may exist, but I can't remember meeting any).
I think the combination of a terrible name (AI implies this technology can "think" and "understand") and imaginations have really added fuel to the fire. There are two main reasons why this is not really a worry.
(1) The AI we have today is extremely primitive. When most people talk about AI, they are talking about the improvements in neural networks that have happened in the past few years. Neural networks can find optimal statistical associations. They are generally just a logistic regression, many times, stacked vertically and horizontally. Fancier networks generally just reconfigure the network architecture (i.e. RNNs) but they all basically work on the same principle.
They generally work by taking a set of inputs (images, audio, sensor data) and known outputs, then they optimize a set of weights that can best predict that output. That's it. This idea that they will somehow decide it's in their best interest to kill humans is far fetched. I'm much more concerned about a rogue dictator deciding it's in his best interest to nuke us. Andrew Ng once made a comment that we should be about as worried of self-aware AI as we should be about over-crowding on mars.
(2) There are many machine learning tasks, but most generally spit out an answer that has to be carried out by plain old boring code. For example, a model might spit out the best ad to show a user, but code has to fetch and display that ad. Even if AI somehow magically becomes self-aware, it's limited by the code we write. So it may magically output "kill all humans", but the only valid choices are "shoe ads" or "purse ads."
In the case of military robots being controlled by AI, it would be trivial to create out-of-band kill switches.
The bigger worry with AI is job displacement. Although I do question whether many companies can get their act together enough to make products that will actually displace jobs. Sure, google can make incredible virtual assistants using some of the best talent in the world. The job displacement will occur in markets much too small for Google to care about. So it will be up to smaller companies to write the software. In my experience, most of these companies can barely get a basic CRUD web app working, let alone a complex neural network with many highly tuned hyper-parameters.
I know a lot of people have ideological objections to WSL, but from a practical standpoint WSL isn't even very good. I tried it for about two weeks before abandoning it.
First, windows has a terrible terminal emulator. I don't think it's improved since Windows 95. Basic stuff like copy/paste is not intuitive, let alone nice features like tabs. I tried an alternative (cmder I think) and it was OK, but something as important as the terminal emulator should not be an afterthought.
Raw sockets didn't seem to work correctly (or at all). I tried a few network tools and they generally fell flat on their face.
It seems really slow. Maybe it's just my imagination, but sometimes I'd do something as simple as an 'ls' and patiently wait.
There was no GUI support out of the box. I had to setup Xming on the windows side. Again, not super complicated, but it seems like little thought was put into it. I don't need a GUI very often (usually just to display plots I generated), but there should have been more effort.
The goal was to basically have python, R, a C compiler, some networking tools, etc, available when I am in Windows and not have to boot a Linux box for basic things. The quality was just too low and went back to using a combination of VMWare and native windows versions.
Maybe it will get better, but it seems like it's trying to solve a problem most people don't have.
Passwords are basically fingerprint authentication for which you have to put your finger in a location n times.
Good luck. The credit data of basically every American eligible for credit was leaked and nothing was done.
I just don't get it. It's a thermostat. I feel like the nest is Lil' Sebastian and I'm Ben Wyatt.
One of the things that has irritated me on occasion is their use of topic algorithms. I'm fairly certain that searches actually use an algorithm like Latent Dirichlet Allocation when you search. So basically, if I search for bad reviews of Dell the algorithm may choose to accept a word belonging to the same topic (such as negative instead of bad). If you've ever seen your google search bold a term you didn't actually search, that would be LDA at work. Most of the time it's useful.
This is fine for trivial searches, but it gets very irritating when I'm searching for very specific words. Is there a way around it? Maybe. But the hide-all-complexity user interface of google tends to discourage finding it and I just use another search engine.
There was a point in time for which having a membership made sense. This year is going to be my last year of prime. I've actually stopped using Amazon in general for most things in the past year.
Amazon relies heavily on third party sellers these days. It seems like the products from these sellers that are prime eligible just have the shipping costs built into the price. For example, I searched on Amazon "adafruit" (a hobbyist electronics company) and one of the first results is the "Adafruit 328 battery." $18.35 on Amazon, $14.95 on adafruit.com.
For certain items, prime (and Amazon in general) has good deals. If I'm going to by a popular $500 electronic product, Amazon w/prime is usually your best bet. But there are huge categories of products for which Amazon no longer (or never did) makes sense.
I remember an NPR story awhile back that many 3rd party sellers were simply buying products on ebay, marking up the price, re-selling on Amazon, and making a killing. People have been trained to use Amazon even when a simple search can find products cheaper from other sources.
Pantry sucks. The few times I've used it half the cans were severely dented and the deals weren't anything great. Basically Wal-Mart prices.
The prime streaming service is just duplicates of content Netflix and Hulu already offer. I'm guessing they just get the cheapest content (like 90's sitcoms).
Ironically, I've gone back to being a fuddy-duddy and just buying my products in-store. I get what I need and leave. It seems like I have far fewer impulse buys in-person as well.
This is not 100% true. There's a range of how interpretable models are. Decision trees tend to be the easiest to interpret, then linear models, and all the way at the other end of the spectrum are DNNs. But there's been a lot of research lately regarding making DNNs more interpretable as well.
Oops, replied to wrong thread.
This is not 100% true. There's a range of how interpretable models are. Decision trees tend to be the easiest, then linear models, and all the way at the other end of the spectrum are DNNs. But there's been a lot of research lately regarding making DNNs more interpretable.