Nitanshu (pqc/acc)

two things it is doable/achievable if - it's doesn't defy laws of physics - actions towards achieving doesn't incapacitate you rest are just excuses

@distributionat yes those are Tier 1 No no no one would ask for tips in India also it totally depends what are the dynamics of your trip if you wanna explore the heritage and culture i'm p sure there're well guided packages not sure about the price though

@nitanshu Tier 1 = bengaluru, delhi, hyderabad, mumbai, kolkata? How much / day would a helper be expected to cost and is tipping for such services expected

🇮🇳🇮🇳🇮🇳 i want to learn more about India 🥰🥰🥰 I would like to go to India some day even though it is very far. If I go to India do I need to speak Hindi or will everybody understand English in India? #India #Travel

ahh nice this is how mostly agentic researches are being done today but again a bit skeptical on the writing style imo writing is very personalized to everyone has their own signature of embedding the knowledge/learning juice like do u also have a feedback loop where the agents keep learning about how do u write and convey stuff ?

launching rectoandverso, an experiment where i write (very long form) essays the same way i write software. i architect, outline, then fire off a bunch of agents to research, write, and review. after the first draft is done, i iterate with ai line by line, paragraph by paragraph, and sometimes even rearchitect the whole thing until im happy with the end result. ive always found writing to be the best way to learn, with a side benefit of helping others to learn as well. and now i can do this 5-10x faster with agents. my main area of interest these days is frontier bio. the first topic is yamanaka factors. enjoy. rectoandverso.com/yamanaka-facto…

ahh nice this is how mostly agentic researches are being done today but again a bit skeptical on the writing style imo writing is very personalized to everyone has their own signature of embedding the knowledge/learning juice like do u also have a feedback loop where the agents keep learning about how do u write and convey stuff ?

@Lyskey can easily endorse this

Unpopular opinion: Shanghai is more beautiful than Paris. No cap.

@Aarav1656 i just got him off the all JEE, IIT playbook anyways in 2-3 years all this things gonna get normalized

and..... My cousin asks, will I get campus placement of 6LPA after pursuing BTech from SRM haha

zcash

was chatting with my cousin who just completed his 12th and this is what we chatted about - quantum resistant enc algo - semiconductors - efficient voice agents - foundational model architectures i was like, bro take a break

@lmrankhan the best usecase for multi agent orchestration to me yet is just plain research like i just spawn multiple discovery agents with the nice and clean implementation of workflows (discover -> synthesize -> verify) which comes with opus 4.8

A lot of people are talking about running tons of agents, parallel workflows, skills, and orchestration layers. Honestly, for building an app, I've found two coding agents running in async works perfectly fine, Codex for backend and Opus/Claude Code for frontend. Haven't had to use more than that, skills, or complex workflows. The bottleneck is usually figuring out what to build, not how many agents you're running or using any of the advanced workflows. I'm sure there are more advanced things people are doing, but for most MVPs or early stage products, simplicity works

@adidshaft @7VoMercy @telegram ys

@7VoMercy @telegram Looks like the use case for Fizz you were talking about @nitanshu

Request for startup: I want an AI agent that plugs into my @Telegram Groups, Notion & Email I want it to track all conversations in the TG groups and update the related notion fields with fresh context, contact info, etc. Every Web3 Company runs BD/ Partnerships/ Sales on TG

Building an AUTORESEARCH for Cribble to use local AI when the system is Idle. lmk what are the things I should be including/have clear thoughts about: I'm not doing web search for now, but will include it later. --- How it works --- Your notes/code/diffs/highlights/reading trails ↓ local deterministic analysis + local LLM passes ↓ virtual research artifacts and insight cards ↓ you review, ask follow-ups, save/publish only if useful --- for the first version, it should be local workspace research. Examples: > “these two notes are about the same topic but are not linked.” > “this code module and this design note describe the same concept.” > “this file changed recently, but the architecture summary still describes the old behavior.” ...now the question maybe, how is it different from Chat HUD? > Chat is reactive: you ask, it answers. > Autoresearch is proactive: Cribble notices useful relationships or open questions while idle. But it should still be controlled: > runs locally, > waits for idle/thermal/battery-safe conditions, > stores results as virtual artifacts first, > cites source files/lines/commits, > never edits your Markdown unless you explicitly save or approve. For the product, autoresearch is the thing that makes Cribble feel like a second brain instead of just a Markdown reader with AI chat. What else should i consider?

@megabyte0x @browser_use why do this, when claude chrome can use ur authenticated session

@browser_use Tried. Works for many but Instagram. After posting a single post it doesn’t work.

Your agents can bypass logins on any website 🥷 Here's how to use Browser Use Profiles: > Create a profile and start the setup > Sync your local browser to Browser Use Cloud > Spin up a cloud browser with your synced profile Setup once, stay logged in. Try it now ↓🔗

@dcbuilder wait, it's like you're opensourcing yourself XD

turning my personal website and its APIs into an AI native interface, CLI and MCP coming soon

@0xkydo @zmanian wait what XD

LIVE ON POLYMARKET!

X gave me the best birthday present @nitanshu is my new handle now

@ch4r10t33r @panditdhamdhere ohh wow

@nitanshu @panditdhamdhere I can tell tou a story from 2017. Or maybe it must be a documentary. You wont belive it! 😅

If you are web 3 startup & interviewing like this in 2026 then it's ngmi

zcash orchard vulnerability : explain like i'm five.

@mike_connor no incentive for exploiter to participate

I think there is technically a way to determine whether the zcash bug has been exploited, _without_ everyone turnstiling out of the orchard pool. (Achieving it _practically_ is... unlikely, but it's interesting...). Unlike turnstiling, the approach wouldn't require users to access their secure spend authorising key (ask), which is important because zcash holders might be reluctant to expose this key. (Classic cold storage opsec). It can be done without revealing the proper nullifiers of any notes. The idea: All users execute a new circuit which proves ownership of an unspent note (as at the block height at which the bug was patched). We sum the values that are outputted by all these circuits. (There are approaches to summing the values without revealing any of the values, but there are tradeoffs here. More on that below). If the sum ever exceeds the amount in the pool, we know the bug was exploited. The circuit: For the block height at which the bug was patched: - Here is a note. - It exists in the tree. - Here is the nullifier secret key (nk) which belongs to the owner of the note. (Notice: we don't require the `ask`, so the circuit never sees enough info to be able to _spend_ someone's notes). - Compute the proper nullifier. - Prove the note is unspent. We don't want to reveal the proper nullifier. Instead, prove the proper nullifier has not been emitted up to the block height. (This requires us to merkleise all historic nullifiers that were emitted from orchard, and record the resulting "nullifier tree root", so that we can do a non-membership proof of this proper nullifier). - Compute an alternative nullifier. We'll expose this, to prevent this note from being counted twice in our summation. The derivation can be the same as the proper nullifier, just with a new domain separator. - Commit to the value of the note, somehow. (The approach here will depend on how we want to sum the notes. Pedersen commitments? MPC secret shares? FHE?) - Public inputs: note tree root, nullifier tree root, alternative nullifier, commitment to the value of the note. Submit the (proof, public inputs) to some verifier. The verifier (which might be a single actor or an MPC network, depending on how we want to securely sum the values) does the following: - Verify the proof - Ensure the roots are correct. - Ensure the alternative commitment is not a duplicate. - Sum the committed note's value. It would be really cool if the values can be summed without leaking any values to anyone. There are MPC protocols whereby a single boolean statement can be learned: "The sum of all values is/isn't greater than the amount held in the shielded pool" I.e. "The bug has/hasn't been exploited". If everyone participates, and if no bug was exploited, the total should match the amount contained in the pool: Once the sum hits the tvl of the pool, if the count of existing (notes - nullifiers) (i.e. the count of unspent notes) exceeds the count of proofs submitted, then we know the bug has been exploited. If _not_ everyone participates, there might still be hope: If the bug _was_ exploited, and if the attacker has already _withdrawn_ material value, we can arrive at our answer faster and _without_ participation from everyone. We'll hit the tvl before we've summed all honest participants' notes (because some of the tvl would have already been rugged). That's compelling, right? Practical problems: - People might not want to participate. - People who've already withdrawn might not care to participate. - People might have lost their keys or notes, so even if everyone wanted to participate, not all notes can participate. But even so, if there has already been a significant amount withdrawn by an attacker, with sufficient participation (not everyone), we could still learn about the attack's existence. Maybe it's simpler to just have people turnstile out. Interesting, though.

@namanc19 @colosseum @incubator we don't need second YC we already have @alliance on Solana

what is closest to YC on solana? @colosseum or @incubator?