I agree that C++ new features have improved it. But many features also feel bolted to me. For example, I prefer Rust's language level support for move semantics than C++'s library level support. Smart pointers are very nice. But Boehm GC integrated into Nim does not seem to add much overhead. And Rust's model integrates smart pointers into the language. But if C++ features feel natural to you, then more power to you. I would have sided with C++ since it has more libraries. But Rust has enough crates now that it does not lag behind too far, at least for my purposes. Nim's packages can't match C++, but it integrates with C/C++ well since transpiles to them. I avoid low-level use of threads and prefer parallel maps and parallel blocks. They suit my purposes, but are perhaps not adequate for you.
Even with C++17, C++ is the other extreme. There are several other languages that fall nicely in between - almost as fast and efficient as C++, while being as productive as Python. If you memory consumption is an issue I would suggest Nim (like Python), Crystal (like Ruby) or Rust (like ML). They are all quite high level and use very tight data structures and consume little memory.
> The amount of overhead on every variable and every member of a list or dict is absurd. There are packages like numpy to help with this, but they're not universally applicable and more limited in capabilities than stl structures in C++.
Then you Cython the heavily used data structures. But more likely, Python is perhaps not optimal for your use cases. In the end, it is still a glue language. But most of the programs typically written in Python never hit these performance road blocks.
...which is when you are supposed to profile to identify your performance bottle necks and write those in a native extension. Python was meant to be used with occasional supplementation by a fast programming language, not replace those. That is how most use it. Python even chose to sacrifice true multi-threading to simplify writing native extensions. If you have a lot of number crunching, you use a GPU math library from Python, not code straight in it.
That aside, the main reason I use Python today is the ecosystem. I started using Python almost 2 decades ago, when it wasn't that popular - back then, I had to explain that it was just a nicer Perl, to people who had never heard of Python. Back then, the high performance languages just weren't very productive to use. Today though, things have changed. There are several programming languages that are almost as performant as C/C++, and yet are very easy to use with several high-productivity features. My favorite is Nim. Rust and Scala are also quite nice. But it is the libraries, not the language that still keeps me in Python. They cover just about everything. Most of the time, performance is the least of my concerns. Time to produce is (research analytics - quick disposable programs).
C takes a huge amount of experience before you stop shooting yourself in the foot and start being productive. Languages like Go are trivial to learn. Rust is less trivial.
It is the Python version of Crystal. It transpiles to C/C++/Node. Has type inference, integrates Boehm GC, good FFI, meta-programming etc. Similar performance to C (CPU, RAM, static binary size), but with the productivity of Python.
Crystal seems to have more modules - both seem to have the essential libraries covered.
> all these 'deep learning algorithms' will always fall short of whatever expectations you might have
Actually, they have exceeded expectations so far,
None of the actual researchers who work on these worry about these going sentient at this level, only the Gates and Musk types.
Actual researchers simply see these as slightly deeper statistical learning systems and are just concerned about building models with better classification accuracy. That itself is tremendously powerful TODAY to keep working on them.
A percentage, without the context of use, is meaningless.
They might be using them for screening, to focus human evaluation. If so, that means that it is ultimately the cop that makes the decision, not the system. This is how today's AI is meant to be used - as a cognitive aid.
It is fairly common for screening tests in medicine to have high false positive rates. That is OK. They are just meant to narrow down the search space for more expensive/invasive confirmatory tests. Given that the incidence of criminal targets will always be a tiny percent of the corpus, it is very difficult to have tests with high true positive rate. That is quite normal for general tests, in general.
The questions that are relevant are:
1. Are the police able to better solve crime with the aids? 2. Is the test too expensive for the said improvement? 3. What are the rates of negative outcomes (like a wrongful arrest) and.. 4. What do we, as a society, consider to be acceptable thresholds?
> I'm trying to figure out why a hyper-rich guy is not only shopping for a bit of electronics himself, but spending time hunting for bargains like a regular broke-ass millennial.
If he is a young exec, chances are that he was a thrifty grad student not too long ago. Most people's spending habits and general behaviors don't automatically get rewired the moment they get a larger pay check.
I still peek at the same deal sites as I did 15 years ago, often just by habit. My income certainly changed a lot in the meantime.
I wish someone makes an alternate language for the Go ecosystem. I like Go libraries: high-level, scripting grade API. I just don't care for the Go language itself. There is an alternate language called Have, but it does not look interesting either. I would like to see something like Nim for the Go ecosystem, or perhaps a Go backend for Nim.
Getting a PhD is very little about coding, even with a CS PhD. Expecting so would be the equivalent of expecting a novelist to be as fast as a professional typist.
No one gets a PhD to become a superior coder, nor do they mostly even care. Math and Physics PhDs typically code better than an average coder, only because they are generally intelligent to begin with, not much because of any training and experience they received while doing their PhD. Of course, they cannot out-code a professional in a competitive position, with decades of experience in just coding.
The CS PhDs I know will solve (practical) problems barely even understood by an average coder. Don't dismiss them because they can't tune a JVM or grok a new Javascript framework as quickly.
And I would not say that a Bachelors in CS does not help a coder. College formally teaches foundations for which there is little incentive to learn on your own if you are too market focused. For those that lack it, the deficit will always insidiously effect the quality of work.
> Yes, he is a bit extreme, but then again he needs to be.
Each time, it is repeatedly shown, that his seemingly extremist ideas simply appear so, only because they are ahead of their time (or rather, most of us are behind time when it comes to understanding current technology). He is far better able to project into the future, what the natural consequences of the current systems are.
Seriously? This is recent? In Seattle, no less? I have seen 100 Mbps in semi-rural India for less than this. That place also has over 20 tiny ISPs competing... all over telephone poles and such.
> When it comes to driving cars they sometimes forget to brake for people
So do people, with greater frequency.
There will never be a perfect AI. Well-surpassing average, imperfect people is advance enough.
Given that the new AI rush started about 6 years ago, where we stand is pretty impressive. Progress may stagnate hereafter, until the next leap, but the advances are undeniable and practical.
They have a new comic bundle out as I type this. They are inexpensive and tend to be eclectic; so not a bad place to start since you are just starting out and do not have specific preferences yet.
I feel the paper editions are too pricey these days, especially when bought individually. Digital comics read well on tablets. I think it is a good idea to stick to digital until you find a series you really like.
You should visit your local library as well and sample around. Mine offers both paper and digital versions.
> So by "modern" we're talking about 90s technology rebranded for the web age?
No, we are not.
> Because while the old delphi component model was convenient
We are not talking about the "old" Delphi. We are talking about modern Delphi.
Once again: take a look at some video tutorial. 30 min will give you a reasonably good idea of what it is. You think Dart is a "proprietry (sic) moon language". It is BSD.
You seem to have used VCL, but not much of FMX. FMX has better fundamentals for a fragmented computing platform era and has better data binding ideas, but got the initial bad rep since the implementation was not up to the mark early on. So you probably are viewing things in that light. BTW, I do not use Delphi ATM but might pick it up again in the future. Yes, it has its own flaws like everything else.
Flutter has MVC and React concepts as well.
> Or is this one of those things like NoSQL where marketers drag out bad ideas we abandoned in the 80s and 90s
NoSQL is good for what it is. Yes, it was oversold and poorly-defined. I have built apps where NoSQL was significantly better performing (this was for read-only data where a relational architecture would involve too many lookups. The data was updated in bulk very infrequently, but needed very quick responses). I benchmarked it against an RDBMS for my scenario before choosing it. You choose by metrics. NoSQL is such a loosely defined term, that we can mean very different things from it It is foolish to put graph data in a relational model, for instance.
Yes, don't jump thoughtlessly into the bandwagon, but don't fear new things either.
I actually would argue that we have taken a step back with modern development. Native RAD that Delphi offered is all the more important for the mobile era. Yet, we devolved into this: Let's pack an entire, separate bloated web browser engine into the simplest web app kind of paradigm. The native options are too platform specific.
How many good cross-platform, native compiling, GPU utilizing frameworks that look good without you needing to be a designer or spending way too much time to get layouts right? How many of them don't bog you down with slow compiles (Delphi with FMX was horrible at this when targeting Android).
Delphi and C++ Builder did good RAD because Borland tweaked the language for the needs of the framework. Google is using Dart in the same way. This article is about some new features they introduced to make Flutter code look leaner.
Of course, modern tools unsurprisingly do some things better than the original Delphi. Why would they not? It has been over 20 years since that model. But they also surprisingly do a lot of things worse.
I have kept up. Watch any short tutorial on Flutter on Youtube. It is not a complicated framework to understand. It is quite different from ReactJS and other hybrid approaches. It is conceptually closer to Embarcadero's (Delphi) proprietary FireMonkey framework. Both use rendered widgets with a GPU accelerated backend.
I should add: The killer framework for Dart is Flutter.
Flutter is the most modern cross-platform mobile framework yet.
- The API is very high level (component oriented). Very easy to pick up. - The tooling is the best there is. Hot reload is fantastic. Only WYSIWYG editors are missing, but UI design is just editing component trees. - Not a Javascript hybrid framework. - Native compilation. - Talks to the platform API, although a bit clunky on that front. - Google's framework (This may be a con for some). - Free.
Cons: - Still not 1.0 yet. - Uses its own rendered widgets that look very native, but could fall behind if Google does not keep up. - 8 MB runtime with slightly slower startup compared to native - probably no different from other third party frameworks like Xamarin.
No one is using Dart to run inside a browser anymore. That idea predictably met the same fate as ActiveX or even Applets.
People are however using it as:
- A better Typescript to generate Javascript.
- A better node.js on the server side.
- A Java/Kotlin/Objective C/Swift alternative for writing cross-platform mobile apps.
It is fine for what it does. There is hardly any learning curve. Think of it as Google's Java/C#, because Oracle was giving them grief over Java.
Slashdot Mirror
I agree that C++ new features have improved it. But many features also feel bolted to me. For example, I prefer Rust's language level support for move semantics than C++'s library level support. Smart pointers are very nice. But Boehm GC integrated into Nim does not seem to add much overhead. And Rust's model integrates smart pointers into the language. But if C++ features feel natural to you, then more power to you. I would have sided with C++ since it has more libraries. But Rust has enough crates now that it does not lag behind too far, at least for my purposes. Nim's packages can't match C++, but it integrates with C/C++ well since transpiles to them. I avoid low-level use of threads and prefer parallel maps and parallel blocks. They suit my purposes, but are perhaps not adequate for you.
Even with C++17, C++ is the other extreme. There are several other languages that fall nicely in between - almost as fast and efficient as C++, while being as productive as Python. If you memory consumption is an issue I would suggest Nim (like Python), Crystal (like Ruby) or Rust (like ML). They are all quite high level and use very tight data structures and consume little memory.
https://github.com/kostya/benc...
> The amount of overhead on every variable and every member of a list or dict is absurd. There are packages like numpy to help with this, but they're not universally applicable and more limited in capabilities than stl structures in C++.
Then you Cython the heavily used data structures. But more likely, Python is perhaps not optimal for your use cases. In the end, it is still a glue language. But most of the programs typically written in Python never hit these performance road blocks.
...which is when you are supposed to profile to identify your performance bottle necks and write those in a native extension. Python was meant to be used with occasional supplementation by a fast programming language, not replace those. That is how most use it. Python even chose to sacrifice true multi-threading to simplify writing native extensions. If you have a lot of number crunching, you use a GPU math library from Python, not code straight in it.
That aside, the main reason I use Python today is the ecosystem. I started using Python almost 2 decades ago, when it wasn't that popular - back then, I had to explain that it was just a nicer Perl, to people who had never heard of Python. Back then, the high performance languages just weren't very productive to use. Today though, things have changed. There are several programming languages that are almost as performant as C/C++, and yet are very easy to use with several high-productivity features. My favorite is Nim. Rust and Scala are also quite nice. But it is the libraries, not the language that still keeps me in Python. They cover just about everything. Most of the time, performance is the least of my concerns. Time to produce is (research analytics - quick disposable programs).
Peter Higgs says he would not have survived in this system.
https://www.theguardian.com/sc...
C takes a huge amount of experience before you stop shooting yourself in the foot and start being productive. Languages like Go are trivial to learn. Rust is less trivial.
No one is expecting C to disappear any time soon.
You should also check out Nim.
https://nim-lang.org/
It is the Python version of Crystal. It transpiles to C/C++/Node. Has type inference, integrates Boehm GC, good FFI, meta-programming etc. Similar performance to C (CPU, RAM, static binary size), but with the productivity of Python.
Crystal seems to have more modules - both seem to have the essential libraries covered.
> all these 'deep learning algorithms' will always fall short of whatever expectations you might have
Actually, they have exceeded expectations so far,
None of the actual researchers who work on these worry about these going sentient at this level, only the Gates and Musk types.
Actual researchers simply see these as slightly deeper statistical learning systems and are just concerned about building models with better classification accuracy. That itself is tremendously powerful TODAY to keep working on them.
Yes, but that was a TV broadcast, not concurrent streaming.
This is an entirely different scaling challenge.
How good is ARM Linux battery life on a Chromebook, if I replace the Chrome OS completely?
Is it comparable or is it much lower?
A percentage, without the context of use, is meaningless.
They might be using them for screening, to focus human evaluation. If so, that means that it is ultimately the cop that makes the decision, not the system. This is how today's AI is meant to be used - as a cognitive aid.
It is fairly common for screening tests in medicine to have high false positive rates. That is OK. They are just meant to narrow down the search space for more expensive/invasive confirmatory tests. Given that the incidence of criminal targets will always be a tiny percent of the corpus, it is very difficult to have tests with high true positive rate. That is quite normal for general tests, in general.
The questions that are relevant are:
1. Are the police able to better solve crime with the aids?
2. Is the test too expensive for the said improvement?
3. What are the rates of negative outcomes (like a wrongful arrest) and..
4. What do we, as a society, consider to be acceptable thresholds?
> I'm trying to figure out why a hyper-rich guy is not only shopping for a bit of electronics himself, but spending time hunting for bargains like a regular broke-ass millennial.
If he is a young exec, chances are that he was a thrifty grad student not too long ago. Most people's spending habits and general behaviors don't automatically get rewired the moment they get a larger pay check.
I still peek at the same deal sites as I did 15 years ago, often just by habit. My income certainly changed a lot in the meantime.
Never mind. The just released GraalVM seems to be everything I was looking for.
I wish someone makes an alternate language for the Go ecosystem. I like Go libraries: high-level, scripting grade API. I just don't care for the Go language itself. There is an alternate language called Have, but it does not look interesting either. I would like to see something like Nim for the Go ecosystem, or perhaps a Go backend for Nim.
Getting a PhD is very little about coding, even with a CS PhD. Expecting so would be the equivalent of expecting a novelist to be as fast as a professional typist.
No one gets a PhD to become a superior coder, nor do they mostly even care. Math and Physics PhDs typically code better than an average coder, only because they are generally intelligent to begin with, not much because of any training and experience they received while doing their PhD. Of course, they cannot out-code a professional in a competitive position, with decades of experience in just coding.
The CS PhDs I know will solve (practical) problems barely even understood by an average coder. Don't dismiss them because they can't tune a JVM or grok a new Javascript framework as quickly.
And I would not say that a Bachelors in CS does not help a coder. College formally teaches foundations for which there is little incentive to learn on your own if you are too market focused. For those that lack it, the deficit will always insidiously effect the quality of work.
> Yes, he is a bit extreme, but then again he needs to be.
Each time, it is repeatedly shown, that his seemingly extremist ideas simply appear so, only because they are ahead of their time (or rather, most of us are behind time when it comes to understanding current technology). He is far better able to project into the future, what the natural consequences of the current systems are.
Yes, or more specifically:
https://www.cpubenchmark.net/c...
https://www.videocardbenchmark...
Google acquired Gizmo project, an open SIP Skype alternative, back in 2009. That was also scanning the whole computer for some reason.
Seriously? This is recent? In Seattle, no less?
I have seen 100 Mbps in semi-rural India for less than this.
That place also has over 20 tiny ISPs competing... all over telephone poles and such.
> When it comes to driving cars they sometimes forget to brake for people
So do people, with greater frequency.
There will never be a perfect AI. Well-surpassing average, imperfect people is advance enough.
Given that the new AI rush started about 6 years ago, where we stand is pretty impressive. Progress may stagnate hereafter, until the next leap, but the advances are undeniable and practical.
You could try Humble Bundle.
They have a new comic bundle out as I type this. They are inexpensive and tend to be eclectic; so not a bad place to start since you are just starting out and do not have specific preferences yet.
I feel the paper editions are too pricey these days, especially when bought individually. Digital comics read well on tablets. I think it is a good idea to stick to digital until you find a series you really like.
You should visit your local library as well and sample around. Mine offers both paper and digital versions.
> So by "modern" we're talking about 90s technology rebranded for the web age?
No, we are not.
> Because while the old delphi component model was convenient
We are not talking about the "old" Delphi. We are talking about modern Delphi.
Once again: take a look at some video tutorial. 30 min will give you a reasonably good idea of what it is. You think Dart is a "proprietry (sic) moon language". It is BSD.
You seem to have used VCL, but not much of FMX. FMX has better fundamentals for a fragmented computing platform era and has better data binding ideas, but got the initial bad rep since the implementation was not up to the mark early on. So you probably are viewing things in that light. BTW, I do not use Delphi ATM but might pick it up again in the future. Yes, it has its own flaws like everything else.
Flutter has MVC and React concepts as well.
> Or is this one of those things like NoSQL where marketers drag out bad ideas we abandoned in the 80s and 90s
NoSQL is good for what it is. Yes, it was oversold and poorly-defined. I have built apps where NoSQL was significantly better performing (this was for read-only data where a relational architecture would involve too many lookups. The data was updated in bulk very infrequently, but needed very quick responses). I benchmarked it against an RDBMS for my scenario before choosing it. You choose by metrics. NoSQL is such a loosely defined term, that we can mean very different things from it It is foolish to put graph data in a relational model, for instance.
Yes, don't jump thoughtlessly into the bandwagon, but don't fear new things either.
I actually would argue that we have taken a step back with modern development. Native RAD that Delphi offered is all the more important for the mobile era. Yet, we devolved into this: Let's pack an entire, separate bloated web browser engine into the simplest web app kind of paradigm. The native options are too platform specific.
How many good cross-platform, native compiling, GPU utilizing frameworks that look good without you needing to be a designer or spending way too much time to get layouts right? How many of them don't bog you down with slow compiles (Delphi with FMX was horrible at this when targeting Android).
Delphi and C++ Builder did good RAD because Borland tweaked the language for the needs of the framework. Google is using Dart in the same way. This article is about some new features they introduced to make Flutter code look leaner.
Of course, modern tools unsurprisingly do some things better than the original Delphi. Why would they not? It has been over 20 years since that model. But they also surprisingly do a lot of things worse.
I have kept up.
Watch any short tutorial on Flutter on Youtube. It is not a complicated framework to understand.
It is quite different from ReactJS and other hybrid approaches.
It is conceptually closer to Embarcadero's (Delphi) proprietary FireMonkey framework. Both use rendered widgets with a GPU accelerated backend.
I should add: The killer framework for Dart is Flutter.
Flutter is the most modern cross-platform mobile framework yet.
- The API is very high level (component oriented). Very easy to pick up.
- The tooling is the best there is. Hot reload is fantastic. Only WYSIWYG editors are missing, but UI design is just editing component trees.
- Not a Javascript hybrid framework.
- Native compilation.
- Talks to the platform API, although a bit clunky on that front.
- Google's framework (This may be a con for some).
- Free.
Cons:
- Still not 1.0 yet.
- Uses its own rendered widgets that look very native, but could fall behind if Google does not keep up.
- 8 MB runtime with slightly slower startup compared to native - probably no different from other third party frameworks like Xamarin.
No one is using Dart to run inside a browser anymore. That idea predictably met the same fate as ActiveX or even Applets.
People are however using it as:
- A better Typescript to generate Javascript.
- A better node.js on the server side.
- A Java/Kotlin/Objective C/Swift alternative for writing cross-platform mobile apps.
It is fine for what it does. There is hardly any learning curve. Think of it as Google's Java/C#, because Oracle was giving them grief over Java.