←
Arpit Gupta67,612 posts
Arpit Gupta avatar

Arpit Gupta

@arpitrage

In DC Sep-Dec 2026 Associate Professor of Finance, @NYUStern. Finance/Real Estate/Urban newsletter: arpitrage.substack.com.

◷ Joined September 2015▧ 9,114 Photos and videos
Arpit Gupta avatar
Arpit Gupta@arpitrage·Oct 1

🚨New predoc position!🚨 @AlexBartik and I are hiring a predoc at NYU Stern, starting Fall 2027 You would help us build AI tools to measure zoning rules and study how regulation impacts housing supply Ideal for AI enthusiasts who want to get a PhD apply.interfolio.com/194498

Séb Krier avatar
⟳ Arpit Gupta retweeted
Séb Krier@sebkrier·2h

Some quick reactions to the recent Dwarkesh podcast with Paul Christiano. Tl,dr: bullish on rapid capability gains, more skeptical of the leap to takeover, and wary of emergency overreach.

Chris Blattman avatar
⟳ Arpit Gupta retweeted
Chris Blattman@cblatts·12h

Nine months is a lifetime in AI. After becoming hopelessly outdated, Claude Blattman is newly refreshed! My motto: Agentic AI for professionals who don't code The idea is to get it to write first drafts in your voice, triage your inbox, manage project teams, and fix your oven.

Post attachment
Arpit Gupta avatar
Arpit Gupta@arpitrage·17h

Removing the tax incentive to borrow: would help efforts to restrain AI development, might slow the financial impacts of a data center collapse, and would help fiscally

“The AI buildout is one of the biggest economic bets in US history, and the tax code is encouraging companies to make it with borrowed money.” Read more from Thomas Brosy & @elenaspatel: tpc.io/48rI9uv

Jacob Schaal avatar
⟳ Arpit Gupta retweeted
Jacob Schaal@FutureEconJacob·Nov 16

New NBER paper: @danbjork, Lagakos & VanVuren write the theory I was missing on whether AI growth reaches developing countries. In @ReadTransformer (@ShakeelHashim), I argued for giving poorer countries a stake in the AI industry. They find the share of AI output needed to stop the consumption gap widening stays roughly stable (20–40%) across AI scenarios, because transfers grow with the size of the disruption. The caveat: this assumes AI’s value shows up as AI-sector profits. If competition drives AI prices down, the endowment pays out little. Here's the paper nber.org/system/files/workin… and my rough sketch transformernews.ai/p/distrib…

Post attachment
Derek Thompson avatar
⟳ Arpit Gupta retweeted
Derek Thompson@DKThomp·19h

Why It Could Get Even Worse For Nvidia It’s not just that hyperscalers are building their own proprietary chips. It’s also that, as @alexbhturnbull has argued, advanced AI is getting better at looking at any piece of hardware and writing optimized machine code for it, in a way that weakens Nvidia’s most important moat, which is its CUDA software ecosystem. My conclusion: “CUDA is one of the greatest achievements of human expertise in the history of computing: decades of code built and refined by human hands. If AI models learn to write superior code for any chip, Nvidia will become the latest student of the bitter lesson, and the company whose chips proved that mass compute can beat narrow craftsmanship would become the lesson’s most ironic victim.” open.substack.com/pub/derekt…

fofr avatar
⟳ Arpit Gupta retweeted
fofr@fofrAI·20h

I asked Gemini 4 Carbon to crack an unread 1466 Sforza Renaissance cipher. Unaided, it used its vision and research capabilities to decipher the text. Gemini: - transcribed 15th-century cipher characters from manuscript scans - ran blind statistical decoding attempts - pivoted to searching digitized European archives for clues - found a handwritten flyleaf chart in Paris showing how Milan’s cipher office built syllable tables - widened its search to all Sforza keys and spotted an unindexed gap in Vienna Codex 2398 - read Tranchedino’s 15th-century Latin index in Vienna and found Giovanni Bianco’s 1465 key on folio 47r - mapped the key signs to decode all 18 encrypted lines in Italian - uncovered warnings of Savoyard reprisals after an ambush on the new Duke of Milan, plus Louis XI’s secret three-marriage plan Full details: fofr.ai/bianco-1466-cipher

Post attachmentPost attachmentPost attachmentPost attachment
Arpit Gupta avatar
Arpit Gupta@arpitrage·21h

It's tough to get a base rate of cyber incidents. Here is one vibe coded attempt, which shows the main break was in 2020-21 and it's stayed there. We are in a dense moment now, but comparable to Oct-Nov 23.

Post attachment

I am beginning to think shit just goes down all the time and we tend to ignore it because we aren't primed to panic that its the beginning of something bigger as we are now with AI

Arpit Gupta avatar
Arpit Gupta@arpitrage·23h

There are domains which will benefit from better intelligence • scientific progress • adversarial domains like cybersecurity, law, or finance • healthcare/drug discovery but in many others: existing models are saturate tasks

Arpit Gupta avatar
Arpit Gupta@arpitrage·23h

In some cases frontier models may be valuable to help orchestrate lower tier models. But @stevehou points out their market share remains very low. What is going to be the economic value of new frontier models? Versus just diffusing existing models through open source?

1/ Today we’re publishing the Databricks AI Index, showing trends from enterprises who manage AI with our Unity Gateway. Data covers >10,000 companies with broad representation (Digital Natives, Fortune 500, Global 2000, etc...). Enterprise AI consumption is changing quite a bit:

Post attachment
Arpit Gupta avatar
Arpit Gupta@arpitrage·Nov 17

It's a three way race between cyber incidents triggering regulatory oversight; RSI scaling expanding model capabilities; and the AI "bubble" popping or not to determine financial resources for scaling