←
Pranay Anchuri59 posts
Pranay Anchuri avatar✓

Pranay Anchuri ✓

@pranaytej

Protocol researcher | Scientist @Offchain | Most recently at: @PrincetonCITP, @xmtp_ | PhD in CS @csatrpi | Views are mine.

◷ Joined June 2009▧ 4 Photos and videos
Ross.pengu avatar
⟳ Pranay Anchuri retweeted
Ross.pengu@_rossstephenson·Nov 8

I am so proud to have been able to lead anincredible group of Emcees for Devcon VII this year in India! ♥️ - 30 Emcees! - From > 20 different countries! - Allocated across 4 days of Devcon! - Emceeing for 7 different stages! - For 100s of speaker sessions! The BIGGEST and most DIVERSE group of Emcees for any single crypto conference in the world! Coming together to become the GLUE that holds Devcon together 🔥 From technical talks, to intense lightning stages, to hands on workshops where anything could happen! Theese Emcees conducted themselves with grace and professionalism! Thank you all for trusting me as your Lead and becoming the EMCEES OF DEVCON! @abena_eth @aledecentral1 @alexanderlee314 @buzea200 @binji_x @0xChiranjeev @chrishobcroft @d_gusakov @DmytroMatviiv @mijovic988 @eshpvt @dutraguss @huvoliveira @JamesNewell @jiminleex @JorkLeonhardt @jgattermayer @kirkthebaird @louviq @MartinBreiten @MissPurrple_ @OfirEliasi @ottokekalainen @pranaytej @robdotrego @zemse_ @solangegueiros @tech_mingler @yq_acc @zaryab_eth @EFDevcon

Post attachmentPost attachment
Yehuda Lindell avatar
⟳ Pranay Anchuri retweeted
Yehuda Lindell@LindellYehuda·Nov 2

This is not the consensus amongst cryptographers - sure, there might be some who think this, but to say that ECC is *likely* to break is definitely not the consensus. There is absolutely no evidence to support this statement. Likewise, to say that symmetric is more secure.

Post attachment
Ariel Gabizon avatar
⟳ Pranay Anchuri retweeted
Ariel Gabizon@rel_zeta_tech·Oct 30

Recent attacks on Ledger supply chain and Binance reinforce the almost obvious fact: Complexity is the number one enemy of real world security, and simplicity its number one friend. This is why as opposed to the narratives, elliptic curve cryptography with its homomorphisms circumventing a multitude of details arising in lattice and hash based signatures and snarks, is actually the most secure.

Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 14

7/ Studies like ours opens a way to ask broader questions about the limits of predictability and the role of predictive models in high-stakes settings.

Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 14

6/ However, some outcomes were much more predictable than others. These are typically related to physical features. Parental outcomes were relatively more predictable than youth outcomes.

Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 14

5/ So, did the ML complexity help? Comparing the winning pipelines with the simple regressors, we found that the median R^2 improved by only 0.01. Even for the relatively more predictable outcomes, the gains are quite modest.

Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 14

4/ The primary result: most outcomes were difficult to predict. The median OOS R^2 is 0.03. For nearly 30% of the outcomes, the best ML pipeline performed no better than a simple baseline.

Post attachment
Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 14

3/ The expanded study allowed us to look at patterns in predictability. We used information from birth through Age 9 to predict outcomes at Age 15. The data has 15k features and we trained millions of models.

Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 14

2/ First, why hundreds of outcomes? The earlier Fragile Families Challenge brought hundreds of teams to predict SIX outcomes in a common task framework. We expanded it to 484 outcomes spanning education, finance, health, relationships pnas.org/doi/10.1073/pnas.19…

Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 14

1/ How well can ML predict people's futures? Our new paper examines the predictability of 484 outcomes. To our knowledge, the first high-throughput study of the predictability of life outcomes. I worked on this during my time at Princeton @PrincetonCITP. A thread 🧵

Akaki Mamageishvili🇺🇦 avatar
⟳ Pranay Anchuri retweeted
Akaki Mamageishvili🇺🇦@kakia1989·Oct 11

We study a question of incentivizing attention in optimistic rollup environment, which can also be generalized to wider ecosystems: arxiv.org/abs/2610.12220 There is a natural separation of lightweight (attention) and heavyweight (validation) roles, which is modeled as a double Tullock contest with corresponding reward functions. Players can choose their roles and how many Sybils they want to deploy for each role, each Sybil costing them extra constant depending on the role. For non-decreasing total rewards for attention, we obtain a necessary condition that allows having a target number of attention players. Decreasing total rewards allow arbitrary number of target attention players, but sounds otherwise controversial. The paper is co-authored with @pranaytej,@BenBerg07141321 and Matteo Campanelli (all from @Offchain).

Pranay Anchuri avatar
Pranay Anchuri@pranaytej·Oct 11

Great few days in London for AFT. Interesting conversations, good friends, and new connections. Ended it with watching Chelsea at the Bridge. Blue is the colour, football is the game, Arbitrum is the chain. 💙

Post attachmentPost attachment
Eli Ben-Sasson | Starknet.io avatar
⟳ Pranay Anchuri retweeted
Eli Ben-Sasson | Starknet.io@EliBenSasson·Sep 28

I like Shafi's advice very much, and I completely 100% agree with her advice to mathematicians in general. AI is changing the way mathematicians work on Math problems. This is a fact, and as shocking and uncomfortable as it is, mathematicians should acknowledge it.

1:21
30:33
Ed Felten avatar
⟳ Pranay Anchuri retweeted
Ed Felten@EdFelten·Sep 24

Fast Feed + PGA, now live on @Arbitrum, accelerates two halves of the same loop: Observe → Decide → Bid → Execute Fast Feed streams ordering + tx data during block construction, before the completed feed publishes (Observe -> Decide). PGA, priority gas auctions determines ordering (priority fee, two rounds per 250ms block), (Bid -> Execute). Important notes: Fast Feed can't frontrun. By the time a tx is published, the sequencer has already fixed its position. Think of FF as low latency observation, not insertion. PGA, similar to timeboost previously, only allows backrunning (not frontrunning or sandwiching) since the mempool is still private. Excited to finally see this in action on mainnet.

Post attachment