Prithvir

look, i’m not a luddite. claude and gpt are my most-used tools every day. but when the industry’s largest supplier calls compute “revenue” in an essay describing lease guarantees and residual-value commitments used to finance customer demand, my antennae go up. who remembers “eyeballs are revenue,” from 1999? and when i say i remember, i mean from books and articles. i certainly wasn’t conscious then...

ideas are so fragile
designers first
then engineering
then customers
then growth
then investors
then bankers
then lawyers
each only after the last has completely bought in
bring anyone in out of order and the idea collapses

the energy infusion from working with people who really really care is unmatched
no peptide, GLP-1, or NAD+ will ever come close

Every time I find a workflow that LLMs can’t handle, the solution is apparently to
*checks notes*
wait three weeks for a new model to phantasmagorically acquire that exact capability.

How many years until the first robot is legally emancipated from its owner?

How many years until TK’s Atoms acquires Uber?

If your physical AI founder doesn’t have alarmingly tendentious opinions about obscure sci-fi novels, it’s time to update your resume.

humanoids today feel like L1s in 2020
fat valuations
abundant capital
real technical progress
but de minimis revenue
humanoids are not commercially useful. we need:
+ fixed morphology
+ bounded environments
+ deterministic tasks
+ closed-loop perception and control
and the pitch can't be “cheaper humans.” it needs to be machines that operate beyond human limits:
+ continuous uptime
+ sub-millimeter precision
+ millisecond reaction
+ mass coordination
+ hostile environments
humanoids are L1s
specialized machines are stablecoins.
generality captures imagination
constraint creates value.

Has anyone built an OpenRouter or DeFiLlama equivalent for robotics, autonomous systems, and defense?
Real-time dashboard tracking funding, contracts, deployments, technical benchmarks, etc.

The neolab trade maps to alt L1s: high-beta proxies on the dominant narrative, rerating faster than the underlying because supply of investable exposure is constrained.
Bare metal maps to BTC miners circa 2019-21: operating leverage on a commodity input, providing convexity on the way up.
The key is knowing which inning we're in.

open source models feel like lewis carroll's red queen.
no matter how fast they run, they’re still running in place. moonshot's kimi 2.6 is a fantastic model. it would be the best in the world...
if only it had shipped in january.

i've given codex full access to all my documents and folders

the labor market is not ready for what codex is becoming
today i gave codex an api key and it:
> pulled 100k rows
> figured out the endpoints
> cleaned the data
> answered questions that would’ve taken a $250k data scientist 2-3 business days
> sanity checked it
it did this in 2-3 minutes
btw these were not simple sql queries
messy questions where it was unclear what formula to use, what data was missing, and how to iterate based on prior experience working directly with me
could see the unhobbling, CoT, and mechanistic interpretability all working in tandem
absolutely fantastic
2026 is incre

Anyone paying attention can tell that, within a year or two, Kalshi and Poly will generate more revenue from perps than from prediction markets