been seeing so many GTA scenes, physics simulations & motion graphics built with Claude, incredible stuff.
but this one comes from the heart โฅ๏ธ
a small technological glimpse into India's deeply rooted spirituality. ๐ฑ
starting with Mahakal in 3D walk around, discover its story, experience the aarti.
Link below ๐
A train ride from Goa. 50 hour Flights across continents. 4 days of unstoppable energy in Mumbai.
@EFDevcon was living proof that Ethereum is much wider than any single blockchain movement. It is a shared pilgrimage of builders, researchers, and dreamers from every corner of the planet.
A complete Devcon 8 recap reel. ๐งต๐
Announcing Generalised Superintelligence! (GSI) ๐
GSI has achieved 250+ novel discoveries and breakthroughs across eight domains, including Quantum, Superconductivity, Chip Design and Mathematics through its Meta Solver Platform. GSI is extremely compute efficient compared to other frontier labs and excels at finding the hard structure in a problem first, before attempting it.
Everyone can contribute to scientific breakthroughs using the kind of superintelligence GSI is building.
Research preview out now at https://t.co/LSwU67gczE
Day 4 of @EFDevcon - awe inspiring builders bringing dreams to life and our very own @DreamStarterXYZ booth ๐ฆโจ
Neurotech empowering paralyzed patients to communicate, global south money rails, autonomous agent swarms, the infinite garden in bloom, and the next DevCon location revealed!
Closing out an unforgettable 4 days with the #Devcon crew!
Day 3 of @EFDevcon in Mumbai was all about culture โจ
Live music on stage, artists designing game characters, builders shipping those games the same day, and a room full of people chanting for Devcon 9 in Africa.
Shot by the @DreamStarterXYZ crew ๐
Day 2 of @EFDevcon in Mumbai was all about the builders โจ
Gaming hubs, lunar rovers, identity, security, and a lot of Indian founders asking the right questions about regulation.
Shot by the @DreamStarterXYZ crew. Full highlights below ๐
after fixing that QuantLib NaN, getting a bit more into the NaNs, saw they're everywhere when options are hours or seconds from expiry.
tested ~40k cases and 182k real market quotes. textbook implied vol solvers failed on 58% of real same-day quotes, autodiff got only 9% of expiry limits right, and py_vollib's gamma was NaN in every expiry/zero-vol case.
so i've started building a compiler that generates pricing code with proofs over the whole input range: no NaN or Inf, and a certified error bound on every Greek.
wrote about what broke, what held up, and where my tool still loses :)
this option pricing rabbit hole led to my first QuantLib PR!
while testing implied vol implementations against high-precision reference values, found one of QuantLib's approximations returning NaN near the money due to numerical cancellation.
fixed it, and the PR got merged! โ
https://t.co/LCT6SS3Zcv
glad i got to meet @ariutokintumi at @EFDevcon! a fellow privacy hawk haha. really cool work on QRAC!
what interests me about EIP-8141 for privacy is the gas abstraction. a privacy pool could pay for its own withdrawals, no need to pre-fund a fresh wallet or rely on a relayer.
the tricky part is fitting ZK verification within the public mempool's 100k gas cap. would love to explore ways around this!
Created https://t.co/HoFotGW6hC a Post Quantum Resistant Accounts for @ethereum now (no ECDSA!) ready for EIP-8141 without EOA submitting txs. Don't protect your EOA keys anymore, those will become useless soon ๐ฅณ @ETHGlobal Mumbai ๐ฎ๐ณ 100% frontend โ
๐ https://t.co/1CHaU4USkm
this option pricing rabbit hole led to my first QuantLib PR!
while testing implied vol implementations against high-precision reference values, found one of QuantLib's approximations returning NaN near the money due to numerical cancellation.
fixed it, and the PR got merged! โ
https://t.co/LCT6SS3Zcv
worked with precision issues in cryptographic computations during my master's thesis.
Currently while reading about option pricing, came across a benchmark quoting prices to 16 digits and started wondering how many of those digits are actually right.
so I computed mathematically guaranteed upper and lower bounds. 3 of the published American put values fall below the lower bound. Honestly the benchmark targets 1e-4 relative accuracy and these are off by ~1e-5, well within it, but errors that size can grow when you compute Greeks like gamma by bumping, and they matter if you use the values as ground truth to test a more accurate method
Article link below ๐