The Big Idea

Error converting content: marked is not a function

  - ### Prolog
  collapsed:: true
- Date: 2017?
  - TODO get the actual date for when I wrote the idea from the video recorded.
- **Chain of thoughts coming together**
  - I have thought out crediting a mini and round camera drone that fits in your pocket. You throw it out and it starts flying around you and following your recording a video. Useful to record daily bites, travel logs, adventures shootouts etc. Toy but meaningful, at least for me, I like recoding daily bites when I walk to work or to my commute, it’s always a pleasure going back to watching those, and re-feel those moments. It’ll be cool and is analogous to popularity of Snapchat, so I’m sure it’ll be the teens who early adopt this toy until it becomes mainstream and commonplace for adults to try on. In all, we will have a lot of video data from our time for the future machine learning algorithms
  - In my thoughts towards what shall one to for humanity, which is difficult question from an existential perspective for how do you what you do does not harm anything, especially now as I got to understand the ugliness of left social justice movement, I concluded that no matter what, education is important, for I can only come to think about these such at a deeper level and uncover my own prejudices and unconscious inclination through education. Education makes you aware, and awareness is not bad.
  - Now, as I was reading historical texts in russels western philosophy book, I am goosebumped to think **how important history is to understand the present**, at least if you want to have an intellectual understanding of the present. Also, as per vrao, history is the way to live a longer or you can say the immortal life.
  - Connecting the above three dots - the toy makes such history available for the future to replicate; replicate? Yes, through VR or anything else that can virtually take you back in time to make you feel and see things as it were. I’m imagining a google earth for history, where instead of zooming in on just geography, **you zoom in on both Time and place**
- **Update oct 2018**
  - Looked around drone building
  - Lots of diy resources available
  - Few companies have pocket sized drones available now. They all have 4 propellers and are noisy. Really low battery life. Not Apple quality. Not even round.
  - Gonna focus of film motion AI
  - Camera that can record and change focus based on what is going on
  - Triggered by watching dev d pardesi song. I wanna record myself but then I want the camera to focus on the table is I pound it hard, focus on my hand, zoom out back to me etc
  - I feel by the time I build the AI tech, we will have hardware update to drones and battery life
  - Recording modes - directors cut (lots of pane in and out, interesting stuff), office (record like office style)
  - Getting started 
		  	1. Camera detects motion or stillness. Comes close on stillness 
		  	2. Rules based camera movements based on UFC101 dataset
  - Big audience on youtube. A starting custom model usecase for cooking videos
		  	1. pointing with finger to an object zooms the camera 
		  	2. —
- **On AI**
	  collapsed:: true
  - AI is a fancy decision tree. A deeply nested and complex algo that could not have been hand codes or would have taken an enormous amount of hours to come up with. This is what I had facied programming since BASIC days.
  - What is your name? 
			  Sid 
			  
			  Hi, Sid! 
			  
			  What is your name? 
			  Anoop 
			  
			  Hi Popat! 
			  
			  #BOOM
  - Deep learning builds abstractions (imagenet model) - layer by layer. These anstractions can be re-used to train additional layer. Revolutionary transfer learning. A big breakthrogh would be narrative learning and decision making. The narrative AI hmm
- ### Timeline
- **2019**
	  collapsed:: true
  - ### jan 
		  
		  i am awesome - the snitch 
		  kathy sierra badass user book on product is how i think about a product 
		  
		  reading master algo book has really elevated my understanding about machine learning 
		  i am now hopeful about the narrative ai 
		  
		  i have come to wccept that most interesting probelems now  wont be solved by knowledge workers , ml has taken over, ml has eaten software 
		  be a fsrmer, let plumbing aid your fsrming 
		  
		  some interesting models and datasets 
		  https://github.com/tensorflow/models
		  
		  todo
		  
		  boy boy boy
  - “If you squint, you can see captioning as a way of radically compressing an image. One of the projects I’ve long wanted to create is a camera that runs captioning at one frame per second, and then writes each one out as a series of lines in a log file. That would create a very simplistic story of what the camera sees over time, I think of it as a narrative sensor.”
			  
			  from
			  https://petewarden.com/2018/10/16/will-compression-be-machine-learnings-killer-app/
			  
			  show and tell https://github.com/tensorflow/models/tree/master/research/im2txt
			  
			  audioset 
			  https://github.com/tensorflow/models/tree/master/research/audioset
			  
			  objct detection tutorial 
			  https://www.edureka.co/blog/tensorflow-object-detection-tutorial/
			  
			  v interesting 
			  from object detection to interaction 
			  https://towardsdatascience.com/how-to-build-a-gesture-controlled-web-based-game-using-tensorflow-object-detection-api-587fb7e0f907
			  
			  some refernces on hand gensture detction 
			  https://youtu.be/Y6oLbRKwmPk
			  
			  dataset - 20bn-jester
			  https://20bn.com/datasets/jester
			  20 billion hand besture in front of laptop or webcam
			  short vidoe clips 
			  
			  how about we tske short vidoe clips as inputs for classification of whats happening? like love photos, they ocnvey 80% of what a 30 second vidoe would hae at times 
			  
			  the talk below explains how this gesture vidoe dataset is trained using 3d convultion netwrks, which works for vidoe 
			  also, it talks about using pytorch over tensorflow
			  https://youtu.be/keffWSqi67w
			  
			  wohoo
			  the somrhing something dataset 
			  just what i spoke of above - live photos 
			  classifies something doing something 
			  [](https://20bn.com/datasets/something-something/v2)
			  hmm.. the dataset is not open soirce 
			  free for academic use only :( 
			  also, th modle accuracy on leaderboard in anout 60%
			  not great
			  2022 date: [88% accuracy](https://paperswithcode.com/paper/improved-multiscale-vision-transformers-for)
  - ### 2019 feb
  - these models are still doing classification 
			  which means the labels are pr determind 
			  and would have to reqsonably small 
			  less than 500? 
			  i want a similar dataset 
			  on movie scenes 
			  or youtube videos 
			  context aware
			  lets solve one usecase at a time 
			  
			  this company looks like something i want to create 
			  selling a product... smart avataar 
			  + other ml assets like models 
			  https://20bn.com/about
			  “Throw an apple in the air and it will drop to the floor. Wave to someone and they will notice and attend to you. This is clear to humans, it's common sense.” 
			  
			  narrative ai would be byond common sense
  - “Our experience taught us that building an A.I. that can perceive, reason, and interact with humans naturally requires relentless effort to push the frontier of computer vision, especially real-time video understanding. This is what we call a situated A.I. When an A.I. can learn to not just detect objects (the nouns) but also grasp the meaning of actions (the verbs) and understand the nuanced situations we experience, amazing things can happen.”
			  https://link.medium.com/95cPEauXsT
			  
			  
			  narrative ai woul not only grasp the meaning of actions but understand it in much higher context 
			  narratives is how humans generalize 
			  narrative ai will infer stories 
			  
			  20bn has some prevtrained model 
			  https://link.medium.com/NiqiMDqYsT
			  can this be used to generate dataset? 
			  the model predicts camera movements 
			  oooooo
			  
			  ok, there are 10 classes with camera movements 
			  like “Approaching something with your camera” etc
			  this could be uswd to create labels for movie scenes? 
			  using ml to create dataset for ml
			  BOOM 
			  
			  office is the perfect show to star
			  camera is always bumpy, zooms abruptly, mostly on noise from somewhere etc 
			  
			   some quora on this “documentary style” 
			  https://www.quora.com/What-is-the-style-of-the-show-the-office-where-the-camera-always-seems-to-be-moving
			  
			  there are other shows as well 
			  could create video slices of each cut 
			  and remove interview shots or anything thats different 
			  could be done in short time
  - ![IMG_0725.PNG](../assets/IMG_0725_1652508677210_0.PNG) https://www.evernote.com/shard/s229/res/104b775d-a2df-402a-88f9-3e271d9816de Technology/Product is not adopted because it is safer 
		  
		  Tech/Product is not adopted because it is faster
		  Tech/Product is not adopted because it is cheaper
		  Tech/Product is adopted if it creates a better human experience.
		  
		  This reinforces the the Kathy Sierra’s making badass user idea - ‘ I am awesome’ 
		  
		  For the lack of better word, yay the naming problem, I am hung up to use ‘snitch’ to describe the autonomous camera drone, and then it hit me - 
		  How about I actually create a snitch? 
		  Something that runs away from users when they try and grab it, it can avoid obstacle, and still stay in vicinity just to tease enough, and at the same time make an awesome video recording of all the action 
		  The convo with Lindsay (cogito annotation team) really helped me connect the dot. She mentioned there are competition in univs that use some robotic snitch already. 
		  Let’s get Jesse involved in this for gameplay design. 
		  
		  With that, I also want to think of a name for my company. I wanna simply call it ‘Tinkering’ or ‘Trial & Error’ - because a) that’s what ML actually is and b) it triggers the purists and authoritarians, although, I do worry this name can get some negative marketing impression during bad company times - so prone to be made fun of. 
		  I like trial and error - it encompasses the philosophy around software development is the modern age which now even gets academic respect thanks to machine learning. In short, empirical methods to problem solving vs analytical 
		  Then again, I am not one camp or the other, for I truly believe that it’s in the center of analytical and empirical spectrum lies the most interesting solutions to most interesting problems. It’s the mix of knowledge work and machine learning, a mix of plumping with farming. 
		  I might also call it ‘Software Farm’ or Digital Farmers Corp haha 
		      
		  
		  Remember - ML routines are not debuggable 
		  “I discuss these theories in terms of two fundamentally different development styles, the “cathedral” model of most of the commercial world versus the “bazaar” model of the Linux world. I show that these models derive from opposing assumptions” - from cathedral and bazaar 
		  
		  Snitch development model - core + bazaar ?
		  
		  “In fact, I think Linus’s cleverest and most consequential hack was not the construction of the Linux kernel itself, but rather his invention of the Linux development model. When I expressed this opinion in his presence once, he smiled and quietly repeated something he has often said: “I’m basically a very lazy person who likes to get credit for things other people actually do.” Lazy like a fox. Or, as Robert Heinlein famously wrote of one of his characters, too lazy to fail. 
		  In retrospect, one precedent for the methods and success of Linux can be seen in the development of the GNU Emacs Lisp library and Lisp code archives. In contrast to the cathedral-building style of the Emacs C core and most other GNU tools, the evolution of the Lisp code pool was fluid and very user-driven. Ideas and prototype modes were often rewritten three or four times before reaching a stable final form. And loosely-coupled collaborations enabled by the Internet, a la Linux, were frequent.
		  ..
		  
		  And the development of VC succeeded because, unlike Emacs itself, Emacs Lisp code could go through release/ test/ improve generations very quickly. 
		  
		  The Emacs story is not unique. There have been other software products with a two-level architecture and a two-tier user community that combined a cathedral-mode core and a bazaar-mode toolbox. One such is MATLAB, a commercial data-analysis and visualization tool. Users of MATLAB and other products with a similar structure invariably report that the action, the ferment, the innovation mostly takes place in the open part of the tool where a large and varied community can tinker with it.
		  ..
		  
		  If the overriding objective was for users to see as few bugs as possible, why then you’d only release a version every six months (or less often), and work like a dog on debugging between releases. The Emacs C core was developed this way. The Lisp library, in effect, was not —because there were active Lisp archives outside the FSF’s control, where you could go to find new and development code versions independently of Emacs’s release cycle.Note 5“
		  
		  just call it, snitch dev kit
		  
		  BRilliAnt
  - ### march 2019
		  came across this term “memes as marker of time” 
		  the video will not only mark but also let one relive those times
  - idea 
		  create a website for machine learning 
		  its the future 
		  and now everybody can try 
		  as farming is more visible and approachable than invisible plumbing 
		  name it eith farming word 
		  softwarefarm.com
		  
		  use deeplearn.js for lts of cool web demos 
		  like https://teachablemachine.withgoogle.com/
		  
		  or https://js.tensorflow.org
  - ### april 2019
  - Phew, it’s been a while. I have been interviewing all around for my next gig. The gig where this dream will manifest and come to life :)
  - Learned quite a internet-ty things from this kid teraching AI coding on youtube - this buzfeed titled you tube was fun to watch - https://youtu.be/NzmoPqte4V4 - Lots of tools covered - like ai logo maker, mailchimp landing page, transfer learning, tensor flow serving
  - Found out about Google Vision Kit AIY (assemble it yourself) - rasberry pie connected to camera to tinker with vision ML models! Yay! What timing! Also, amazing to see such ML kits pop up from big Cos. Like Amazon has a race car, and maybe even a vision kit themselves. This will at lest let me start with tuning a vision model on video! Maybe even zoom in and out on scenes.. man what a good start it’s gonna be. Can’t wait. https://aiyprojects.withgoogle.com/vision/
  - oct 2019
  - Alltagsgeschichte—” the history of everyday life”
			  “In Germany in particular, this trend has culminated in the practice of Alltagsgeschichte—” the history of everyday life”—achieved through a “thick description” of the common experiences of ordinary people. When such an approach has been applied to the era of the Third Reich, however, some have criticized it as an evasion—a way to shift attention from the unparalleled horrors of the Nazi regime’s genocidal policies to those mundane aspects of life that continued relatively undisturbed. Thus, the very attempt to write a case study or microhistory of a single battalion might seem undesirable to some.”
  - Dec 2019
  - From “words that change minds” book 
			  “From Noam Chomsky and many others, we know that people do not actually live in Reality. By deleting, distorting and generalizing, we inhabit our perceptions and interpretations of Reality.”
			  “Let's say a person has an experience. When that person talks about his experience, they only communicate a minute portion of the actual event. They have to edit out the vast majority of what was going on, just to be able to communicate it in a reasonable time frame.
			  Sid - Boy It all comes down to story telling
			  The narrative human or the narrative AI 
			  So the question - what is delete, what we distort, and what we generalize
- **2020**
	  collapsed:: true
  - Feb
  - Malang
  - March 
		  
		  Avidya
		  
		  Root vid = knowledge 
		  
		  Latin - vid for vision 
		  
		  vision = knowing
  - ### april 2020
  - https://www.csail.mit.edu/research/controlling-drones-and-other-robots-gestures
  - lots of drone activity for coronavirus social distancing measures
  - ![IMG_3247.PNG](../assets/IMG_3247_1652508717265_0.PNG)
  - ### may 2020
  - ![IMG_1574.PNG](../assets/IMG_1574_1652508603073_0.PNG)
  - “The general principle here is: come up with more of a "spread" in your assumptions about humans in forecasting. Don't anticipate the future of the world for just your tribe (or favorite enemy tribe) 44/ If you only have heroes, add villains. If you only have heroes and villains, add in-betweeners. If you have a spectrum, add a dimension or a distribution of populations. 45/ If you, like me, lean foxy, with stronger imagination than nerves, you'll notice thinking with a richer range of human types evokes scenarios that get on your nerves more. 46/ Reality, as Philip K. Dick noted, is that which doesn't go away when you stop believing in it. People who don't go away when you stop believing in them are a big part of reality.”
			  
			  
			  
			  “It is easier to invent the future than to predict it (Alan Kay) 
			  
			  A good science fiction story should be able to predict not the automobile but the traffic jam (Fred Pohl) 
			  
			  Technology and science [are] primarily cultural carrier bag[s] (Ursula LeGuin, see here10) 
			  
			  Any sufficiently advanced technology is indistinguishable from magic (Clarke's 3rd law) 
			  
			  The two hazards of prophecy: failure of nerve, failure of imagination (Clarke) 
			  
			  Speaking of the fourth one, incidentally, check out this great application of the idea of advanced technology as magic: Rodney Brooks' The Seven Deadly Sins of Predicting the Future of AI11.”
			  
			  // 
			  
			  People looking awkward taking snap in public 
			  
			  Snitch could help? 
			  
			  https://twitter.com/influencersitw
- **2022**
	  collapsed:: true
  - #0 Inbox
  - check out [Diffgram](https://github.com/diffgram/diffgram)
  - ### May
		  collapsed:: true
  - Been a while. 2 year hiatus from researching snitch 
			  Let’s blame it on compass🙂 
			  
			  What I have come to believe is that if I can focus on a problem, I will solve it. 
			  Just browsing through the notes here gives me joy. I’m really into deep understanding and on a path. I have found gems from all across to put the puzzle together. 
			  
			  I know I have unraveled the greatest mirage of human mind - the self
			  Narrative AI could be sort of solved in 15-20 years
			  going by apple timeframe apple computer to iphone 
			  jobs vision was always the iphone- a personal computer for consumer; my mom has it
  - why, how, why
    - why: to create historical artifacts that’s as is - real, personal, without **interpretation**
    - how: creating companion recorder, a personal **art** director for you
    - what: snitch; flying camera that can **capture a story**
    - “the goal is to do business with people who believe what you believe” - simon sinek. Law of diffusion of innovation.
  - ### June
		  collapsed:: true
  - Jun 23rd, 2022
    - **05:03** quick capture:  https://www.linkedin.com/posts/chrisgpresents_seo-google-tiktok-activity-6945462721656086528-ukH0?utm_source=linkedin_share&utm_medium=ios_app #The Big Idea
  - Jun 26th, 2022
    - **06:52** quick capture:  https://milvus.io/blog/scalable-and-blazing-fast-similarity-search-with-milvus-vector-database.md
    - 06:54** quick capture:  https://github.com/milvus-io/bootcamp
    - ****06:56** quick capture:  {{twitter https://twitter.com/memdotai/status/1539391178461749248?s=12&t=-0iscM3k2cJrQT0aHINSOA}}
  - **July**
		  collapsed:: true
  - drone running on computer vision blowing everyone's mind with its speed and precision
			  id:: 62c983ed-adfd-4361-a403-f3b62a2c8fd8
			  collapsed:: true
    - **07:32** quick capture:  {{twitter https://twitter.com/wholemarsblog/status/1545120731406614528?s=12&t=rTmf7K9c6dtzKXW7NLBSzQ}}
  - Wow! really good Research Tools
    - https//www.Elicit.org
    - ![image.png](../assets/image_1658276681319_0.png)
    - Connectedpapers.com
- 2023
	  collapsed:: true
  - Jan
		  collapsed:: true
  - Heard the use word 'Narrative' the first time. In Narrative BI Ad on twitter - narrative.bi
    - ![image.png](../assets/image_1673102344349_0.png)
  - - How the future is shaping towards this vision
- Came is the eye. Video is the reality.
- 2017 - tiktok launched
- Rise of #videotech since 2018
  - failure of @quibi - holywood meeting, vertical video
- 2021
  - Apple announced Cinematic Camera Mode with #iphone13
  - Cinematic mode works by **capturing a depth map of the video as it's being filmed**. In essence, capturing both the foreground, midground and background of the scene for it to be able to apply various focuses or blurring effects beyond what is capable via the lenses without additional help.
  - TikTok Overtakes Facebook for Screen Time
  - TODO https://tech.co/news/tiktok-overtakes-facebook-screen-time
  - TODO https://www.data.ai/en/go/state-of-mobile-2021/