Prathmesh Vankore

Made it to the Hall of Fame 🔥

@anandlunia @IndiaQuotient Super pumped to see you there!

IndiaQuotient team is coming to #Mumbai on 9th April to meet early-stage founders. Looking forward to it! luma.com/p7su09t0

@jay_kotakone @kotak811 Hey Jay Tremendous respect for what you and the team are building at @kotak811 🔥 Always impressed by the audacious thinking and how far ahead you guys seem to be building! Kudos 🫡

9 years ago, we made a bet that most people thought was too ambitious. Other industries had already figured out that convenience was the product. Banking had perhaps missed the memo. So we built @Kotak811 like a startup. Obsessively product & customer first. The complete banking suite, accounts, cards, loans, insurance, investments, UPI & more, all on your phone. All digital. All yours. Powered by the trust of Kotak Mahindra Bank. Today, that bet has manifested as India’s most downloaded banking app & over 25 million customers. We’re just getting started. Happy Birthday Kotak 811. 9 years in, ♾️ to go.

Prediction markets always broke on Parlays Too many siloed combinations, shallow liquidity ParlayMarket: avoids 2^M explosion → M² parameters fragmented liquidity → 1 shared pool All joint bets draw from the same pool with O(M²) scaling Very interesting!

@DStrachman Our cows are saying hello to your bunnies :) (just emailed you)

During the day, fund manager extraordinaire. By night, supreme bunny bonding handler. My home is basically a petting zoo now.

It’s not 10,000 hours, it’s 10,000 iterations.

@refsrc China had to solve supply and scale first wherease India and LATAM are still solving trust, consistency, and last mile delivery more of human problems.

Paul Graham on why you shouldn’t write with AI: “In preindustrial times most people's jobs made them strong. Now if you want to be strong, you work out. So there are still strong people, but only those who choose to be. It will be the same with writing. 𝐓𝐡𝐞𝐫𝐞 𝐰𝐢𝐥𝐥 𝐬𝐭𝐢𝐥𝐥 𝐛𝐞 𝐬𝐦𝐚𝐫𝐭 𝐩𝐞𝐨𝐩𝐥𝐞, 𝐛𝐮𝐭 𝐨𝐧𝐥𝐲 𝐭𝐡𝐨𝐬𝐞 𝐰𝐡𝐨 𝐜𝐡𝐨𝐨𝐬𝐞 𝐭𝐨 𝐛𝐞."

@harshilmathur does UPI AutoPay via Razorpay support non-cyclic, on-demand debits? Or do we use the subscription mandate itself for flexible, non-recurring payments? Would love to get some clarity here 🙏

@Tim_Org Moral hazard of not having true skin in the game.

I also think that the field was hurt by certain incentives: those who wanted to apply the research in organizations and get the consulting fees and stage time, and reporters who were looking for flashy headlines. Good social science probably ought to be kind of boring.

I wrote an essay for Bloomberg Weekend on social psychology. I was curious about where the field stood after all of its scandals: not just fraud and fabrication but results that didn’t replicate and the "desk drawer" problem that hid null results.

Liquidity seems to win the Schelling point game every time in such cases :)

everybody is too worried about the cost of doing the thing and not worried enough about the cost of not doing the thing

@SarkaryShahvir Looks super exciting! Keep posting such magical stuff 👀

i like robots that glow in the dark

@DStrachman Huge shoutout to my younger self for taking cold emailing seriously 🫡

Mixmax, you are fucking with my love of cold emails (see pinned tweet folks) -- just turned off the settings where these messages don't reach me (and might not have for the last week or two) and go to some nebulous unfindable folder. If you sent a cold email in the last week or two, you might try again!

@paraschopra @karpathy True Though novelty-seeking id kind of a human survival hack, just like empathy partly comes from status signalling It'll be interesting to see if AI can pick up that meta-rule by default without some alignment nudges

@Prathmesh2604 @karpathy yeah but same with humans interestingly, many non-verifiable activities are socio-cultural in the sense that the average becomes boring and people crave for now so in that sense, an AI system could learn that meta-principle also and start to self-correct

Sharing an interesting recent conversation on AI's impact on the economy. AI has been compared to various historical precedents: electricity, industrial revolution, etc., I think the strongest analogy is that of AI as a new computing paradigm (Software 2.0) because both are fundamentally about the automation of digital information processing. If you were to forecast the impact of computing on the job market in ~1980s, the most predictive feature of a task/job you'd look at is to what extent the algorithm of it is fixed, i.e. are you just mechanically transforming information according to rote, easy to specify rules (e.g. typing, bookkeeping, human calculators, etc.)? Back then, this was the class of programs that the computing capability of that era allowed us to write (by hand, manually). With AI now, we are able to write new programs that we could never hope to write by hand before. We do it by specifying objectives (e.g. classification accuracy, reward functions), and we search the program space via gradient descent to find neural networks that work well against that objective. This is my Software 2.0 blog post from a while ago. In this new programming paradigm then, the new most predictive feature to look at is verifiability. If a task/job is verifiable, then it is optimizable directly or via reinforcement learning, and a neural net can be trained to work extremely well. It's about to what extent an AI can "practice" something. The environment has to be resettable (you can start a new attempt), efficient (a lot attempts can be made), and rewardable (there is some automated process to reward any specific attempt that was made). The more a task/job is verifiable, the more amenable it is to automation in the new programming paradigm. If it is not verifiable, it has to fall out from neural net magic of generalization fingers crossed, or via weaker means like imitation. This is what's driving the "jagged" frontier of progress in LLMs. Tasks that are verifiable progress rapidly, including possibly beyond the ability of top experts (e.g. math, code, amount of time spent watching videos, anything that looks like puzzles with correct answers), while many others lag by comparison (creative, strategic, tasks that combine real-world knowledge, state, context and common sense). Software 1.0 easily automates what you can specify. Software 2.0 easily automates what you can verify.

@paraschopra @karpathy Interesting Though doesn’t this still risk Goodhart’s Law kicking in?

@karpathy the key point is that many of the non-verifiable domains can be broken down into rubrics, each of which can be verified so i'm bullish that AI will make rapid progress even on softer domains like poetry or film-making

Everyone reading the Gravity Aging Hypothesis by Continue Research. @deepigoyal

Species win by beating Entropy. Innovation creates New order. Coordination makes it Scale.

मुझे खुद से मिलने के लिए, एक पुराने दोस्त से मिलना पड़ता है..। @Prathmesh2604