lonerapier.xyz

Things on Internet that I don't want to forget

20 min read

Other links:

Jun ‘25

May ‘25

  • Uses This / Interviews: Nice interviews of people from different professions about what their current tools are.
  • Jony Ive and Patrick Collison: What a beautiful conversation!!
    • “Having a clear sense of goal which is to enable and inspire people”
    • “Solving a functional imperative, and we’re done. Of course, that’s not enough. That’s not the characteristic of an evolved society.”
    • “To people, Simplicity is about removing clutter. But Simplicity to me, is about succinctly expressing the essence of something, and its purpose, and its role in our lives.”
    • “I think how you feel while working on something gets ultimately embodied into the final product. So if I’m anxious, that’s how end product will end up. I think to be hopeful, and optimistic, and joyful in our practice, and be that way, in how we relate to each other.”
  • Why did DeepMind solve protein folding? - by Jake Feala: Interesting insights on AI for biology, and why protein folding isn’t the singularity point. It’s still an inflection point for biology, and research labs.
  • A Baby Receives the First Customized CRISPR Treatment | TIME
    • “That treatment involves removing cells responsible for generating blood cells from a patient, then genetically editing them using CRISPR to turn on a gene that makes fetal hemoglobin, which is normally turned off in adults. Once the blood stem cells are edited, they are then re-infused back into the patient. The idea is that these cells would start to make more copies of themselves and eventually generate enough healthy red blood cells to minimize or even eliminate the painful symptoms that patients experience.”
    • This is Medicine 2.0
  • “India’s remittance tax woes in Trump’s ‘big, beautiful’ bill”, FT: Immigrant Indians sending 20% of their income to the home country with total amount exceeding $30B in FY23-24.
  • “Has Starlink already won the new space race?”, FT
    • The congestion of LEO: Estimates says 100,000 satellites fighting for space and path in Earth’s most priced space resource.
    • SpaceX alone has launched 39% of total satellites launched since Sputnik, and has approximately 8000 starlink satellites in orbit. It’s already approved for 12,500 satellites.
    • Generates 2B in free cash flow in 2024-25 alone.
    • “Ultimately, it aims to fly more than 40,000 satellites.”
    • “Amazon will need to spend between 20B to build Kuiper, Quilty estimates.”. That’s really expensive for any developing nation to bootstrap. A nation like India that spends ~1B in its entire space program will never even think of such an effort.
  • How Does Claude 4 Think? – Sholto Douglas & Trenton Bricken - YouTube
    • Models solved intelligence problem in last 2 years. Next step is long-term agentic capabilities.
    • Where they struggle in agentic performance? Models can work on small, tightly-scoped task or long widely recognised boilerplate tasks seamlessly. Where they struggle right now is the exploration phase of a loosely-scoped, iteration-heavy task that requires multiple changes across the environment of the agent.
    • What happened with RL from last year? New kid on the block: RL with VR (Verifiable Rewards). RLHF has been the primary technique used to train these LLMs on human tasks. Verifiable Rewards work on scenarios where reward function is deterministic, and objective. While RLHF is analogous to a subjective response. Example: coding use case of unit tests passing, or completing a task on the web.
  • Overcoming India’s technological cowardice
    • Promise → Despair → Hope with a special emphasis on absolutely bizarre state of overregulation, administrative incompetence, and government underfunding.
    • How traditional family run businesses work in India: “Apply trade barriers for world-class companies outside India, create a low cost half-good copycat product for Indian consumers, sell to poverty stricken overburdened consumer, invest the profits in media entertainment to sell to the same exceedingly free and overstimulated viewer.”
    • ISRO’s death story hits much harder.
    • What are the deep tech sectors that need a world class Indian representative on the world stage?
      • Space: Agnikul, Pixxel, Skyroot
      • Autonomous Agents
      • Drones: IG Drones, DroneAcharya
      • Biotech: PopVax
      • Gaming/Entertainment: Nazzara
      • Semiconductor
      • Critical Minerals
      • Batteries
      • Energy
  • Fundamental Development Gap Map v1.0: List of unsolved gaps in R&D.

Apr ‘25

Mar ‘25

Feb’25

  • “Everyone knows your location: tracking myself down through in-app ads”, tim: this is beyond scary. Imagine the data owned by by big organisations, regarding your interests, financial capacity, relationships, geographic location. It’s not just targeted ads, this ad-based economy has been influencing people already. Mega organisations have been accused of altering election outcomes, race hate, propagandist thinking several times now. It’s supposed to end at some point, and I think the arrival of agents will solve that to an extent. Internet will become more AI friendly and less user friendly. Information will be hidden behind paid APIs
  • Will DeepSeek deep-six the US economy?, by Steve Hsu and David P Goldman
  • “The Generalist’s Productivity Stack”, The Generalist: Raycast’s focus extension has done wonders for my focus. It fights that initial friction and agitation, and helps to avoid any and all distractions.
  • “MODERN-DAY ORACLES or BULLSHIT MACHINES?”, Carl T. Bergstrom and Jevin D. West
  • “Terence Tao on how we measure the cosmos | Part 1”: 3b1b goated content never ends. It really blew my mind that initial mathematical intuition began from asking simple questions and using logic + analysis to reach the answer. It didn’t include any sophisticated mathematical equations, but very lengthy process of data collection, observation, and pattern matching through the series of unrelated dots that when connected gave the answer in plain sight.
  • “sparkly people and how to find them”, Anson Yu: I haven’t found many of these people, and most probably, that’s due to my introvert nature. Personally, i have like 3 or 4 of these values. I’ve had to fight for many things to get to where I am, how I learn, and how i perceive things. I won’t deny that it’s not fun, but, not always. Understanding a field deeply, in order to have significant impact, takes time and effort, which you can’t be expressed as having serendipitous fun.
  • “An Interactive Introduction to Fourier Transforms”, Jez Swanson
  • “NEW SPACE” Frontier Film: “We’re explorers, it’s just how we’re programmed, it’s imbued in our genes.”
    • “Why Starship Matters”, Casey Handmer
    • I can’t help but feel so much childlike wonder after watching this. We are going to accomplish these goals, because that’s what humans do. We will form a nuclear-powered permanent base on Moon, and then Mars, along with a thriving civilization by the end of the century, or even before.
    • One another thing that is clear is it just doesn’t matter if you have the perfect solution for a problem, Starship is the most brute-force approach for a space conveyor belt, but it’s the best we have right now. We are obviously going to discover much better solution than this in the future, but you gotta start, and you gotta start “now”.
    • Pairing starship with Starlink satellites was the best outcome, anyone could’ve come up to bootstrap the space cargo economy.
  • “Speed matters”, Jamie Brandon: Moving faster in anything you do has many more indirect consequences. Spending a week trying to learn something, then spending another 2 weeks trying to make a demo out of it, sound cool. But thinking, how can I do it 10x faster, i.e. both of these in 2 days rather than 21 days saves much more time.
    • This is also along the idea of how to read a new research paper. I still read research papers every paragraph front and back. But that’s just a massive waste of time. You just need to read the abstract to know you understand every word that the author is going to say in the paper, this means you skip any introduction and preliminaries, jump straight to actual result, and any evaluation or performance benchmarks. That’s all it needs to read any kind of paper.
      • This arguably takes just 1 day for reading a 100-page paper, than 2 or 3 days to read every line and paragraph of it. It also gives your brain to think about the blanks after you’ve read it, it’s like a hardass problem that you’ve now etched into your subconscious mind to ponder on.
      • I’m trying to learn this method more and more, and applying this with other orthogonal areas can really speed up the process of learning a new thing.
    • Similarly with coding, I think i spend 100x more time trying to find a good solution rather than just code the brute force approach, test it out, and improve later. Testing really gives your brain time to think of each step you wrote, why it’s there, can it be improved, or removed entirely.
  • “50 Years of Travel Tips”, Kevin Kelly: all of these are really good tips.

Jan’25

  • “Putting Ideas into Words”, Paul Graham: Writing thoughts and ideas down formally (heck, even informally) is one of the most therapeutical thing one can do. It’s not easy, you start to uncover many flaws in your understanding. What seemed like a 1500 line essay turns into a 200 line, scattered, unfinished, mostly factually or logically incorrect think train. But, that doesn’t mean you stop writing. It’s the only way I know to slow down my thoughts, and to move them from working memory to long term memory. And it’s not just words, you can do the same thing with code, drawing, cooking. To put more precisely, actually building something physically.
  • “Privacy guides”: Extensive list of privacy guides for browsers, DNS, VPN, cloud providers, etc. Highly recommended.
  • “Machines of Loving Grace”, Dario Amodei: 5/5. well-explained 10K ft view of what AGI will and won’t do. Most of the article can be summarised as being a lengthy discussion on human processes, which might be all true. AGI’s, according to me, will be considered useful, if it can design processes far more efficient, creative and organised than the human counterparts that are mostly riddled with bureaucratic administrative problems. Easy way of estimating what AGI can achieve is to imagine 100 parallel researchers, working on different problems in different domain, but with same accuracy. With speed and efficiency, will also come reduced cost of deployment, and thus, ultimately benefitting humanity with low-cost, accurate, and efficient end products. These products can be substituted with anything: medicines, drugs, therapy. Revisit
    • “All Watched over by Machines of Loving Grace”, Richard Braughtton
    • Is it possible for AI to reduce economic disparity within a society, and internationally? If yes, what’s the path forward? How do we reduce the constraints set by humans?
    • How to use that extra intelligence in reducing the per capita income gap of developing vs developed countries?
    • Reduce poverty → reduce corruption. Goal of every government across nation boundaries should be to lift the economically backward section of your society to a point where it has access to cheap education, healthcare, household.
    • What’s the threshold for BPL in India? How much families have been thrived out of BPL in last 5 years? What were the major regions where these families are located?
  • “Beyond nature and nurture”, David Bessis: I think it’s disrespectful to label someone as untouchable geniuses. There are obviously, people whose mathematical intuition, and imagination is much more capable and develop at an early age, and there are also people who invest every second of their life to reach that step. I personally, like and do mathematics for the problem solving. There are no shortcuts in enjoying it, you have to get your hands and mind dirty.
  • “Things unlearned”, Jamie Brandon: Good list of things that worked for a person doing programming for 10 years. Most of the advice is reoccurring in retrospective pieces that we see flying around a niche part of the internet. Problem solving, intelligence over expertise, mathematical foundations, High leverage options, better time management. I think everyone knows that these are the steps one need to take, it’s the execution that’s difficult. Discipline is what’s most needed here.
  • “my phone is making me dumb”, Isabel: Yes, and a whole lot yes. You know this is true, yet you can’t escape the tentacles of the giant squid that the algorithm has become.
  • “Reading as a creative act”, Bits of Wonder: This is solid advice that is applicable to almost all form of entertainment consumption. Textbooks, fiction or non-fiction books, videos, music, even short form content like essays, or reels. All of these need to be consumed with much greater friction, i.e. each should have some questions that it answered, or created new ones, or helped clear doubts, or helped formed new beliefs. Otherwise, we’re all hamsters on a wheel. I’m also guilty of doing passive reading, even when learning through textbooks. I’m too focused on completing a lecture, or finishing a book, instead of completely understanding what the material wants to convey.
    • I think the solution is to have a list of questions that you want answered from the text, form new questions as you go through it, set of questions that the text was able to answer.
  • “Future of Energy Reading list”, Casey Handmer: TODO
  • “Life Lessons from the First Half-Century of My Career”, David A. Patterson: some that i’m practicing still:
    • “Most of us spend too much time on what is urgent and not enough time on what is important.” - I’m really horrible at time management, and works on anything that I find interesting. While this sounds good, but is really sketchy in practice. For example, while reading a research paper, when stuck on a problem, i’ll distract myself with household chores, that leads to me wasting 2-3 hours, making me run in circles with the initial hard problem that I left.
    • “Look for the positive opportunities.” - I’m a really negative person, confidence comes hard to me. I’ve recently just seen that just labelling a task as easy or doable in x days can decrease the complexity by significant amount. Having amazing people around you to ask for help always helps, but in the end, any task has to be completed by you. So, look for those positive opportunities in a project, and double down without ever looking back. Obviously, retrospect later.
  • “I Ditched the Algorithm for RSS—and You Should Too”, Joey Hand: trying to set up nice RSS, and move over completely from algorithm based timelines.

Dec’24

Nov ‘24

Oct ‘24

September 2024

August 2024

July 2024

June 2024

Graph View