@sundarpichai For an Indian use case, I'd love to see performance broken down by ancestry, cell type and rare variants. Predicting a molecular effect and proving clinical usefulness are different milestones. Both deserve clear benchmarks.
@deepakshenoy@itsTarH Median adds a useful lens alongside aggregate profit growth: a few heavyweights can't dominate it. Next checks: how many companies have reported, the sector mix, and whether growth came from revenue, margins or a low base.
@zerodhamarkets Useful side-by-side comparison. The next step is separating trading-linked revenue from other income, then checking volumes and fee changes. That helps explain a QoQ move without treating exchange earnings as a direct measure of participation.
@Iamsamirarora Start date is the most underrated variable in any market chart. Measure from the Sep '24 peak and India looks broken; look at earnings, flows and breadth since then and you see what actually changed. Price tells you mood, not trend.
@elonmusk Same lesson at small scale: in production AI, moving and syncing data is harder than the model.
Software built to route data across lossy, high-latency links is a real edge for training across clusters. Could India's new GPU clusters tap it?
Under the "Chinese models" memes sits a real production question: which model for which task?
Cheap open weights for bulk extraction and tagging. A frontier model only where an error costs money.
Your monthly AI invoice tells you if you got that split right.
Currently, I have all of the above, and personally, I feel Claude Code and Codex are the best value for money; Cursor consumes many tokens, and my weekly limit exhausts in eight hours of coding, whereas that is not the case with Claude and Codex. Regarding Grok, after adding the all-in-one plan, it is not useful for coding considering the limits.
@deepakshenoy LRS gets blamed for FX pressure because it's the visible retail line.
Trade deficit, FPI and RBI intervention move dollars far more. LRS falling while the rupee stays soft is the tell that the remittance-as-villain story was always too neat.
@elonmusk The AI use case that actually sticks isn't robotaxis for office workers β it's letting a 98-year-old keep his independence.
India's about to hit that same demography wall. Autonomy as care > autonomy as status.
@deepakshenoy@itsTarH Median is the right lens here. Aggregate profit growth gets bent by a few banks and oil cos; the median tells you the typical company actually grew. Next question for me: how much of it came from margins vs revenue? Margin-led beats rarely last.
@elonmusk In India, 'signal everywhere' matters less for video and more for a UPI payment that doesn't fail at a highway dhaba or hill village. Satellite plus towers on one phone could do more for rural payments than any app. Price per GB decides if it reaches them.
@elonmusk The real shift is Starlink going from satellite add-on to a full terrestrial carrier. If that model reaches India, the fight moves from rural coverage to who owns the subscriber, and the ARPU math Jio and Airtel built on spectrum costs.
@zerodhamarkets El NiΓ±o's hit to food prices usually comes with a lag: weak rains, then pulses and vegetables, then rural demand and RBI's room to cut. Rabi sowing and reservoir levels are the early tells to watch, before the CPI print.