Tarun Chitra

@pinged

204 Following

24708 Followers

Yano pfp

Wow, amazing episode. Apollo has $750 billion AUM. Now they’re moving onchain. Finally got Apollo partner and head of digital assets on Empire to discuss their new onchain credit fund and crypto journey. If you’re building or investing in RWAs, this pod with @christinemoy, @pinged, and Carlos Domingo is a must listen.

2 replies

2 recasts

14 reactions

Tarun Chitra pfp

As someone who got back to NYC from Dubai in time for the end of FarCon.. should I do anything? cc @jayhinz https://x.com/jayhinz/status/1918457566226428338?s=46

2 replies

0 recast

14 reactions

Rehash: A Web3 Podcast pfp

Rehash: A Web3 Podcast

For our last episode of the season, we welcome @pinged to the pod! This week, we discuss DeSci, AI in crypto, over-financialization, and longevity research... and the complexities that come with each. Links below:

2 replies

5 recasts

14 reactions

Rehash: A Web3 Podcast pfp

Rehash: A Web3 Podcast

As always, thank you to the Rehash community for nominating and voting on our guests! @pinged was nominated onto the pod by @davidphelps on behalf of Camille Vargas Voters included @treegirl @davidphelps @richie @triumph @anay @andyhyfi.eth and @ernestkou

1 reply

1 recast

5 reactions

Tarun Chitra pfp

I read this one

0 reply

0 recast

5 reactions

Tarun Chitra pfp

4 replies

4 recasts

59 reactions

Tarun Chitra pfp

https://maps.app.goo.gl/kyQovRmXokt5yhJy9?g_st=com.google.maps.preview.copy

0 reply

0 recast

1 reaction

Tarun Chitra pfp

@alok has a photonics PhD 😅

2 replies

0 recast

1 reaction

Tarun Chitra pfp

haha having worked for someone who won one and prior to that we worked for someone else who wished they won it, I'd probably say not chemistry So Physics!

0 reply

0 recast

1 reaction

Tarun Chitra pfp

I think the RWA meta is coming to vaults soon 😀

1 reply

1 recast

3 reactions

Tarun Chitra pfp

Great question; I think there are a lot things to consider: - There's a natural trade-off between isolation and optimality; figuring out scenarios where this is true or not is crucial - Sometimes, you can show that if you can create a marketplace for people competing (like Morpho/Euler) can actually "learn" risk preferences — https://arxiv.org/abs/2503.18237 - A question is if you can synthetically learn this preference (e.g. via AI) in a robust manner (e.g. deals with MEV) — I suspect the "worst case" answer is no, but the average case might be yes

0 reply

0 recast

4 reactions

Tarun Chitra pfp

Probably mainly investing; despite really enjoying using AI right now, I'm not sure I see that much that's long term investible or defensible (beyond M&A, at least); but I assume that might change in the next 6m

0 reply

0 recast

2 reactions

Tarun Chitra pfp

I guess the question is what you need security for; there are definitely applications where this is immediately true (e.g. MEV, pre-confs) which make up the revenue distributing AVSs/networks [e.g. primev] But I'll give you a question: Suppose I enter a SaaS agreement with a particular SLA. The vendor doesn't uphold the SLA; how big of an infraction does it need to be for me to sue them? This is one of the questions you have to ask when assessing the minimum threshold for where you need security; money (e.g. a PoS network) and financial applications (DeFi) obviously needs it for arbitrary sized infractions. Non-financial applications, however, do not seem to have a clear defined "minimum security budget"

0 reply

0 recast

4 reactions

Tarun Chitra pfp

Changes a lot! But for something that has an amazing video aesthetic (and isn't really pure techno) https://www.youtube.com/watch?v=3CCFIMrPpDY&t=2892s

1 reply

0 recast

6 reactions

Tarun Chitra pfp

1. Probably would try to make my hair look like a different Ilana Savdie painting every day 2. Glasses: If its digital, why stick to a single pair? You can have a continuously varying and random family of glasses unlike IRL 3. Would probably force my digital twin to do all the homework problems in Hartshorne that I couldn't do

0 reply

0 recast

1 reaction

Tarun Chitra pfp

Intellectual 1: Can you solve a millennium prize with a human + AI? Intellectual 2: Can we ever fix classical statistical learning theory to deal with LLMs? Does it matter when reasoning seems to be able to correct shittier models and make them "almost" as good as better models? If this is true, doesn't this mean double descent [https://www.pnas.org/doi/10.1073/pnas.1903070116] doesn't exist for reasoning models? Applied Research: How can you make reasoning models 100x cheaper? Practical: Is the AI market completely oligopolistic even in perfect information? If so, doesn't it mean that 90% of startups who raised are actually zombies?

1 reply

0 recast

4 reactions

Tarun Chitra pfp

1. Having Do Kwon and SBF on very close to the end (its like an NFT of a doomer memory, you know) 2. Watching Haseeb go from Solana hater to Kyle Samani maxi 3. Some of our more difficult to deal with guests 😀

1 reply

0 recast

6 reactions

Tarun Chitra pfp

For the first time, I'm weirdly excited about crypto x ai Most of what we've seen so far has been "too early" or "too dumb"; and as they say, "in venture bets, if you're too early, you're wrong" But I generally think the economics of reasoning models becoming less and less galactic makes the opportunity for a competitive market to form around them much more likely than what we've seen so far; DeepSeek was perhaps the first salvo in a long battle to make the cheapest possible reasoning model ever [Wrote a paper indirectly related this, plus some threads on FC; https://arxiv.org/abs/2504.09777]

0 reply

0 recast

5 reactions

Tarun Chitra pfp

Haha, not much to say other than there's ~$700m of happy Gauntlet vault depositors

0 reply

0 recast

0 reaction

Tarun Chitra pfp

I think there's a huge gap between the (current)best programming models (e.g. Gemini 2.5 pro/Claude 3.7) and the models best at math. I think the only certainty is that AI models will be super different in the next year — a year ago, people were still shilling vscode! The OpenAI models are the only ones that seem to not be overfit to contest math evals; you can give them abstract math problems and they can generally parsimoniously figure things out. I'm writing a blog post on this (plus some threads here) that will give more details

0 reply

0 recast

3 reactions