Do Not Write Below This Line When Filing Initial Application: Fill & Download for Free

A Turning Point in HistoryJuly 9th, 2019 is a day to remember and yet it passed by with hardly a notice. It was such an important event and yet it received very little news coverage. It is a day and an event that will probably be forgotten in history - like the first guy that used fire to cook with or the first person that made a wheel.On this date, Lawrence Berkeley National Laboratory reported that they had been conduction experiments with an Artificial Intelligence program they called Word2Vec to sift through scientific papers for connections humans had missed. They used machine learning and deep learning to train the Word2Vec algorithm by having it sift through 3.3 million previously published scientific papers. While doing this, Word2Vec learned the meaning of over 500,000 words.As with most AI programs that uses deep learning, Word2Vec was not trained to find anything specific - only to find patterns, relationships, implications and logical conclusions that can be drawn from the data. AI programs also modify their own programming during the deep learning process in ways that the human programmers and operators cannot follow. Some AI programs can add millions of lines of code as it learns. The operators don’t really know what the AI program has learned until they test it. So they did.Anubhav Jain, a researcher at Lawrence Berkeley National Laboratory, and his team fed the Word2Vec AI program only papers published before 2009 and it was able to predict one of the best modern-day thermoelectric materials four years before it was discovered in 2012. This is an incredible accomplishment. If we had used this software in 2009, we would have made a discovery then, that was not really discovered until 2012. Imagine the cost of the research done in those 3 or 4 years. Imagine the advancement of the sales and benefits of using that newly discovered thermoelectric material 4 years earlier.This was, of course, just a test on an obscure topic but the implications are massive. As noted in the article, dozens and soon hundreds of research facilities are going to obtain this AI program and use it to enhance their own research. Other labs will use it to initiate all new discoveries.At this early stage, significant advances can be made using research papers from past research. In other words, no new lab research will be conducted. New discoveries will be made in hours or days based on just running the software on data files of past studies. Think about that. If 100 labs started using this approach to their research in different fields of study - physics, cosmology, biology, metallurgy, chemistry, etc. - all of these labs might be making ground-breaking discoveries every few weeks or months instead of one or two of them making discoveries every 3 or 4 years.What is also very likely is that other labs and university research facilities will copy, modify and then enhance the Word2Vec AI program so that it can refine its analysis and look deeper into the implications of the studies and data it reads. The next step will be to expand and enhance the AI capability to use it for ongoing research even down to one or two studies.Imagine this. A cure for a disease is being sought. The usual way is to expose the disease germs to a wide variety of possible drugs or substances that might work until one is found that kills the germs. This can sometimes take tens of thousands of trials and years of research. An alternative approach might be to allow the AI program to read all that we know about the disease germs and then to feed it millions of research papers on past studies of drugs, disease, germs, and cures and have the AI program make the connections of treatments that might work. It might, for instance, determine that a two-step cure might be needed. The best-known example is Edward Jenner's use in 1798 of cowpox to prevent smallpox. The advantage, of course, was that cowpox was easily cured but smallpox was not. Since then, dozens of other cases have been found where one disease cures another but all of them were found by accident. Now imagine having an AI program find cures like this and then also using reverse genetics to reduce or eliminate the effects of the second induced disease. This is not something that could be discovered by any researcher in any ethical study but it could be done in hours by an AI program.The benefits, warnings, and threats of AI programs have been discussed for years but it was always off in the future. Now it is here. Now it has been done. It will only explode into the most rapid rate of technological advancement in history and it all started on July 9th, 2019.Below are a few of the 4,710 references I found related to the above referenced research on Google.Text Mining Machines Can Uncover Hidden Scientific KnowledgeAI Trained on Old Scientific Papers Makes Discoveries Humans MissedArtificial Intelligence Set Loose On Old Scientific Papers Discovers Something Humans MissedUsing unsupervised machine learning to uncover hidden scientific knowledgeMachine-learning algorithms can discover new thingsAI Makes New Scientific Discoveries By Analyzing 3.3 Million Scientific Abstracts - Grata SoftwareArtificial Intelligence Read Old Scientific Papers and Made a DiscoveryNew material promises better solar cells (This is one of many in which computers led researchers to new discoveries but this was a dedicated simulation with a specific goal - unlike the open-ended effort of LBNL.)UPDATE #1:This answer has gotten way more attention than I ever thought possible and a surprising amount of it is negative which is a double surprise. Part of the negative comments come from the LBNL choice to use a rather older version of a software called Word2Vec. I suspect this was done because it was free, open-source and was sufficient for their purposes.Details of the inner workings of the software used and in what form the output was produced are secondary and minor details of what was accomplished. There is no doubt that the AI software did not “read” millions of reports, resolve all the connections and language and then wrote a paper of discovery. However, the software they used did “process” millions of reports, established links and patterns in their content and put out a result that was interpreted as a discovery of an as yet undiscovered thermal electric material. The net effect is the same.Other negative comments were that this research is nothing new and is not important. This may be true from the perspective of the level of sophistication of the software but what I was mostly amazed at is the nature of the application and what it allows in the future. Even in this regard, this is not new. Here is an excerpt from a cosmology meta-study of the Hubble Deep Field image taken in 2009 of what was thought to be a void in space but it turned out to contain over 3,000 galaxies.“In general, the fields of astronomy and cosmology have been getting crowded with many more researchers than there are telescopes and labs to support them. Hundreds of scientists in these fields do nothing but comb through the images and data of past collections to find something worth studying. Combining past studies of images with more recent studies using radio telescopes or looking at this new data from these recent examinations of voids has created whole new sets of raw data that can be examined from dozens of different perspectives to find something that all these extra scientists can use to make a name for themselves.”The connection to my answer is that serious and productive research and even valuable discoveries can be made by reviewing data from previous studies and archives of past research papers. This takes on new meaning when we can obtain the synergistic benefit of thousands or millions of such studies looking for patterns, links, connections or interrelated evidence by thousands of researchers, college students, very small labs and commercial companies with nothing more than some easy to acquire software and a computer.Think about it. The LBNL researchers were experts in AI, software, data analysis and deep learning and yet they discovered what would have been a breakthrough in the field of materials science and thermal conductivity. This was way out of their field of expertise and they were not specifically looking for that result. That strikes me as significant.Imagine every university in the world having such capability with tens of thousands of grad students looking for “something” new and undiscovered using only a computer and access to old archived data. This may not be earth-shaking or a paradigm shift in science but it might be the aeolipile that eventually led to the steam engine.UPDATE #2For those that keep saying that AI does not write its own code or create new lines of code, may I suggest the following:An AI can now write its own codeDevelopers, rejoice: Now AI can write code for youNew A.I. application can write its own code - FuturityGoogle's AI can create better machine-learning code than the researchers who made itAI learns to write its own code by stealing from other programsIs Self-modifying code the best option for creating 'human-level' AI?The military just created an AI that learned how to program softwareUsing Artificial Intelligence to Write Self-Modifying/Improving Programs

“Why is C and C++ such a spectacular examples of incompetence?”This is not a question, and you know that very well.You don’t like C/C++, that’s OK. They’re probably not a good match for your daily programming tasks. You probably don’t master these languages. You’re not alone, Quora is full of web/mobile/application developers. They’re poking C with a long stick, wondering what could such a strange thing be useful for, and talking a lot of nonsense. However calling the dark energy of the IT world “incompetent” tells more about you than about technology. You know, all major operating systems, almost all high-performance applications and the complete embedded computing is held together by C. Anywhere you look, below a thin layer of higher level languages, everything is C.C in embedded computingI’m working with automotive in-vehicle software. Mostly bare metal; if there is an OS used in the system, it’s from us, we develop and port it (AUTOSAR/Nucleus/Linux). The industry standard is C here. Control system ECU’s are pure C, ADAS / infotainment is mostly C++.Let me explain in a nutshell, how a typical application works. Execution starts at the reset vector. The first couple of hundred code lines are assembler; first we need to set up the environment for C. Configure and initialize the CPU core, put it into a clean state to avoid Lock Step exceptions when saving state later. Configure the memory management unit to be able to access RAM, rest of the flash and on-chip peripherals. Reconfiguring the TLB entries describing flash often requires to code to be copied to and executed from RAM, this further complicates the startup code. Configure the watchdog to prevent timeout during initialization. Initialize RAM ECC, configure caches. After basic CPU initialization, C initialized variables have to be initialized by copying the init values from flash to RAM. If RAM code is needed (e.g. for flash programming functions), after copying the data cache has to be cleaned, write buffer synchronized and instruction cache invalidated, to force it re-reading the updated code. Call stack, C stack, small data pointer etc… have to be set up, then finally main() can be called.In main(), initialization continues. Interrupts, exceptions have to be set up. Self Test Control Unit, Fault Collection and Control Unit have to be initialized and errors handled properly. CPU run modes have to be initialized, peripheral clocks enabled and configured. PLL has to be set up, timing adjusted for the high clock. Then all peripherals and subsystems have to be initialized.The application typically consists of a main computing loop. First inputs (sensors) are read, then control signals are calculated, finally outputs (actuators) are updated. Diagnostic is usually negligible in runtime, but might contribute >70% of the memory footprint. Dynamic memory makes no sense, as there is only one application and only static objects. In the automotive domain, dynamic memory is explicitly banned by a MISRA rule. It might be needed to switch between modes with different memory layout, then it has to be handled manually by switching between two static memory layouts. Typically using an union; it’s usually needed to avoid violating the strict aliasing rules anyway.Peripherals often generate interrupts to signal service demand. The interrupt function has to save the current CPU state, check the event and dispatch the correct interrupt handler, then restore the context and continue execution of the main thread. Vectored interrupt controllers and good compiler support can help here a lot, however especially with older devices, it’s still needed to use interrupt prologs/epilogs or dispatcher functions written in assembler.Blaming C…Try to do the above in Python. Or Java. Or Rust. Or any other language than C/C++/assembler. Now that would be a spectacular example of incompetence! It has to be understood that C solves different tasks than other languages. That’s why C is different. It’s the interface between electronics and programming, the interface between the hardware and higher level languages. High level languages are created based on philosophies, approaches. C is based on reality: how a CPU core works. But still offers reasonable levels of abstraction, this made it unique and incredibly successful.We need direct memory access. E.g. I need to change the run mode of my PowerPC e200z4d core to DRUN. I open up the datasheet, it says, the value 0x30005AF0 has to be written into the ME_MCTL register, which is at address 0xC3F9C004. So I write the following code:*(volatile unsigned long*)0xC3F9C004uL = 0x30005AF0uL;Or probably I’d define a macro:#define ME_MCTL *(volatile unsigned long*)0xC3F9C004uLOr include the header file from the silicon vendor with all definitions already there. Registers or memory are sometimes located at address zero, so accessing them needs zero casted into a null pointer, and then dereferencing the null pointer. (see Ferenc Valenta's answer to What actually happens when dereferencing a NULL pointer? Usually, the process terminates. Does the reaction depend on the operating system, or is it controlled by the compiler? Is it mandatory that NULL always be defined as “0” with proper casting?) Java guys are already lost, because they don’t know what a pointer is, and only heard about null pointers as an example of spectacular incompetence.Direct interaction with the CPU core often requires issuing special instructions. They’re CPU architecture dependent. It can be done by either inline assembler or using builtin/intrinsic functions, if supported. A further challenge is that the compiler has to be told not to move the critical instructions during optimization, and the CPU core has to be told to conclude the side effects of all preceding instructions before touching the interrupt flag. Enabling/disabling interrupts on ARM Cortex M4 core and GCC:#define barrier() do {__ASM volatile ("dmb");} while(0) #define __enable_interrupt() do {__ASM volatile ("cpsie i" : : : "memory"); barrier(); } while(0) #define __disable_interrupt() do {__ASM volatile ("cpsid i" : : : "memory"); barrier(); } while(0) Or sometimes it’s not a must, just an excellent optimization opportunity. The below example is from an USB driver for Atmel AVR32 UC3C using the IAR compiler. It returns the index of the lowest numbered endpoint requesting interrupt, from the interrupt flags:const uint32 epidx = __count_leading_zeros(__bit_reverse(pending>>AVR32_USBC_EP0INT_OFFSET)); The __count_leading_zeros() and __bit_reverse() are intrinsic functions supported by the compiler and they directly compile into the single-cycle CLZ and BREV assembler instructions. Due to their rather sophisticated function, they’re not automatically generated by any compiler. The alternative, which is often seen in similar applications is a loop traversing through all the bits and checking them one by one with a mask. That’s what you’d implement in high-level languages. With C and it’s direct HW access execution time can be cut down to a small fraction.Besides peripheral access and optimization, you might need to control the memory layout. E.g. interrupt vector table might have to placed to a fixed address or might need special alignment. Or need to change compiler warning level, optimization etc. Then you add a lot of #pragma or __attribute__() to your code. There is an unprecedented freedom in controlling the memory layout, resource usage, program flow, special core features. C supports virtually all hardware, from the smallest Harvard-architecture microcontrollers with 64 bytes of RAM and 1k of code flash to the most complex high-end x86 cores. Not only supports them, but unleashes the full performance of the HW, which is not possible with any other language.All the low-level stuff, which you’d like to isolate you from with boilerplate, are absolutely necessary to solve the C tasks. And the “shortcomings”, you’re whining a lot all the time, like lack of automatic memory management, garbage collection etc… are completely out of scope for the applications of C. They’re not only not useful and discouraged, but sometimes explicitly forbidden. Should you need it, you can code your allocator in C for yourself.SummaryWhy is C and C++ such a spectacular examples of incompetence?They’re not. The question is.

When you start a new project in XCode, it sets up enough to build a working app. For sure, it doesn’t do anything, but it has a window, a menubar, responds to clicks, etc.So that’s a huge confidence builder that everything is working. To get to that stage:File > New > Project. Select OSX/Application and choose Cocoa Application. Next.Give it a name and fill in other info as you wish:Click Next. Choose a location on disk where the project and all its files will be saved and click Create.You’ll get a new IDE window with the starter source files, initial interface files and everything all ready to go:At this stage, resist the temptation to fiddle. Just do Product > Run. It will build the app, interfaces, etc and run it. You’ll get a single empty window and a functional menu bar. Verify it works as far as you can tell.Now you’re up and running, you can make it do more. Add a button to the window for example. The list on the left of the IDE window is all the files that make up your app - source code, headers, interface files, and anything else you add as you go (like images or sounds). Select the file MainMenu.xib in that list, and it will open an Interface Builder editor for that file.The second from left list are all the objects in the interface file. You’ll see objects representing menus and the blank window. Select it, click on ‘Window’:Next drag a Push Button from the list on the right hand side into the window.Choose Product > Run again. You’ll see your app now has the button. You can click it, but it doesn’t do anything yet. Time to write some code.In AppDelegate.m (select it in the IDE), add the following:This defines an action visible to Interface Builder (‘IBAction’) that you want to run when you click the button.Go back and select MainMenu.xib again. Make sure your window and button are visible. Double-click the buttonand change its title to “Beep” (not vital, but it shows how most interface things are editable directly).Right-click on the button and drag out a line to the ‘Delegate’ object icon, shown as a blue cube. Release, and a small window shows listing the method you just added, ‘buttonPressed:’. Click on it to link the button to that action.Choose product > Run. Click the button, hear the beep that you coded.Let’s make sure we can debug. While the app is still running, go back to XCode and select AppDelegate.m. Find the line you added with NSBeep(). Click in the gutter next to that line so that a dark blue arrow shows up. That’s set a breakpoint. Go back to the app and click the button. Xcode will intervene and show you the debugging screen:It has stopped at the breakpoint you just set, highlighting that line. The list on the left shows you the calling stack (stack trace). The panel below shows local variables. There is a small terminal interface to lldb next to it, if you want to use it. There are buttons to continue, single step, step into, step out and so on. Click the continue button to resume execution of your app.That’s how to get started. XCode is huge, Cocoa is huge, getting productive and proficient will take a while, but it’s really just the same steps as here, over and over.