Posts
1. Having bought Discovery+ a year ago, I have been catching up with the shows I was so afraid of speeding through when i bought it first. The main show is "Planet Earth" narrated by David Attenborough. In the initial few episodes, I was blown away by the production value of the show. The show is about every part of our planet earth and how it behaves and functions. If a show is about the entirety of earth, one fundamental problem is getting to all those places. But seeing the footage of all these regions, and the amount of effort it must have taken to film all rare shots, I wonder how much it must have cost BBC to make them. To simply send those people to all these places must have cost a fortune, and considering the risk for life those extreme places would present, the compensation would have to be a lot. Did BBC actually produce those ? Or did they outsource it ? I dont want to know to be honest.
1a. On the topic of Planet Earth, it has to be mentioned that, apart from its generally spectacular visuals and audio, one thing that I absolutely loved was the introduction and the end credits scene in the background. It took me a while to notice. The introduction features Earth from space, where the sun slowly rises from the right side of the screen, slowly illuminating the rest of the screen's background. In the end credits, we see in the background, a visual of earth during sunset, slowly going back to darkness while the credits roll to an end. Such a poetic credits background for a show named "Planet Earth".
2. Part of my job, requires me to get accustomed to C++. But I think I might have fallen into love with it in the process. It is not elegant, and it can be quite verbose most of the times. But everyday when I use it, I discover something that blows my mind. The standard library is soo diverse, that I am of the belief that if I ever find something in C++ that is not part of the standard library, I would almost lose hope instantly in solving that problem. It all started with my introduction to move semantics. My biggest gripes about Python, golang and other languages are, those languages dont make it easy for us to figure out whether an assignment is triggering a copy or just a reference copy. I have spent hours splitting hairs trying to understand why a list of list of numbers change all together when i change one element in the 2D list in python. Also, there is typically only one way an assignment can work in either of these languages. Either you can copy a structure or you can move it. If you want to move a structure, which gets copied by default, Tough luck. But not in C++, Everytime i avoid an unnecessary copy of a heavy structure like a 1000 element vector, I fall one more step closer in love with C++.
3. Have you ever wondered why Unions in C and C++, are even called unions ? It didnt make any sense to me when I learnt it. It is just a bizzare data structure, where all elements occupy the same space. Where would it be useful, except in weird places inside hardware code, where a register can be read multiple ways ? Only 10 years after I encountered it, I finally understood why it is called Unions ! And that is thanks to the typing library in python. You see, in python there are no type restrictions in anything. To make our code look halfway sane, some brilliant people have come up with type hinting the code, so that some IDEs can warn you if you assign a name to a variable which is supposed to hold your roll numbers. But python code being python code, there can be places where we might want to use the same code for multiple types. What do we use in those cases ? Union[type1,type2]. But why ?
Because the final type of the variable is a Union of the two sets of types, and a union here indicates the set operation. In C and C++, the union type allocates a single space for all its types, because the user is expected to use that variable as any one of these types only. The fact that an 64 bit integer can be read as a double inside the same union, is just the side effect of how C handles the memory of Unions. Union types were meant to give us options in types.
3. Have you ever wondered why Unions in C and C++, are even called unions ? It didnt make any sense to me when I learnt it. It is just a bizzare data structure, where all elements occupy the same space. Where would it be useful, except in weird places inside hardware code, where a register can be read multiple ways ? Only 10 years after I encountered it, I finally understood why it is called Unions ! And that is thanks to the typing library in python. You see, in python there are no type restrictions in anything. To make our code look halfway sane, some brilliant people have come up with type hinting the code, so that some IDEs can warn you if you assign a name to a variable which is supposed to hold your roll numbers. But python code being python code, there can be places where we might want to use the same code for multiple types. What do we use in those cases ? Union[type1,type2]. But why ?
Because the final type of the variable is a Union of the two sets of types, and a union here indicates the set operation. In C and C++, the union type allocates a single space for all its types, because the user is expected to use that variable as any one of these types only. The fact that an 64 bit integer can be read as a double inside the same union, is just the side effect of how C handles the memory of Unions. Union types were meant to give us options in types.
4. ChatGPT is all the rage now. But why should it be a chat-bot, that suddenly behaves startlingly intelligent, and not the numerous other equally, if not more freakish AIs ( like DALLE2 and the video equivalents from Google and Facebook ) ? Somewhere I have heard that, scientists believe that LLMs ( Large Language Models ) will be the key to AGIs ( Artificial General Intelligence ), and some even believe AGIs when they arrive, will just be LLMs. If you think about it, that is very weird. Why should a model which predicts the next few words of text, of all things, be so powerful, when there are far more impressive feats that a AI is already achieving, like generating unbelievably real looking fake images. <Opinion> I think it might be because getting anything useful out of text, requires far more than understanding patterns. By trying to solve a very difficult problem, the LLM is forced to understand human language. And human language, especially the one in written form is not very versatile in terms of individual units of information ( alphabets ). So by trying to predict the next word in a sentence, a LLM is expected to understand meaning, language structure. And human language which is limited in versatility, has evolved to communicate complicate stuff like feelings, thoughts, fears etc. These difficult stuff didn't need to evolve in other form of our senses, since they were already very versatile. Since human language is doing such a complicated job with a handicap, it must be extremely information dense compared to other form of senses. </Opinion>
This made me think about limitations in general, and how limitations can lead to beauty in unexpected ways. Limit a language to only use words which rhyme every 5 words once, we got poetry. Limit an instrument to produce one kind of sound, suddenly we get some specific sorts of squeaky bois, giving us chills, making us cry and giving birth to all sorts of emotions. Limit a film director's ability to just say everything that is happening in a movie, we get brilliant pieces of visual story telling. Maybe I am looking at it the wrong way ?
This made me think about limitations in general, and how limitations can lead to beauty in unexpected ways. Limit a language to only use words which rhyme every 5 words once, we got poetry. Limit an instrument to produce one kind of sound, suddenly we get some specific sorts of squeaky bois, giving us chills, making us cry and giving birth to all sorts of emotions. Limit a film director's ability to just say everything that is happening in a movie, we get brilliant pieces of visual story telling. Maybe I am looking at it the wrong way ?
No comments:
Post a Comment