Pradyumna Shome (pradyumna.bsky.social)

Closing the Intel TDX Page Fault Side Channel, or, The Case for TDExit-Notify collective.flashbots.net/t/closing-the-…

(Reposts appreciated!) Do you work on user-facing Artificial Intelligence (AI) products? We would like to hear from you! Please fill out this survey to sign up for the study: cmu.ca1.qualtrics.com/jfe/form/SV_2o…

Calling vibe coders 👾 If you’ve ever used Cursor, Copilot, or AI agents to accelerate your programming in any capacity, join our study! 🎤 60 minute interview (audio recorded) 🎁 $25 Amazon gift card 📋 Must be located in the US and 18+ Sign up: cmu.ca1.qualtrics.com/jfe/form/SV_39…

🚀 Clueso is LIVE on Product Hunt! In just the last 6 months, over 100,000 videos have been created using @goClueso. And now it’s time to take things to the next level! Show us your support and watch the full launch video here: producthunt.com/products/clueso

We've pulled together some of the most interesting research problems in MEV in the new Flashbots Research Problem Database ⚡🤖 It includes completed work, early ideas, WIP, and requests for research in Incentives, cryptography, PBS, TEE, data, and more! flashbots.notion.site/21f6b4a0d87680…

I was today years old when I found out that `const` doesn't protect the underlying container from modification in ES6, just reassignment. Never thought I'd miss good old `const int* const p`!🤣

Operative (@operative_sh) is an AI website creation agent that writes and tests code in your browser — so anyone can build and launch a working web app in minutes. No coding knowledge required. Congrats on the launch, @never_settles_ & @erikqu_! ycombinator.com/launches/Ncy-o…

🎉Excited to announce today that @erikqu_ and I have been accepted into the X25 @ycombinator batch. Erik and I met in 2014 in high school doing coding competitions together. >10 years later, we're building @operative_sh - the world's first web app coding & debugging agent. 🧵

🚨 New Preprint! 🚨 Smarter LLMs are more selfish. We show reasoning-enhanced models significantly prefer greed over cooperation. The more LLMs reason, the worse they cooperate. 👇 THREAD 👇 [Link to paper: arxiv.org/abs/2502.17720] [1/n]

🚨 New Paper! 🚨 "Groups that communicate diverse information make wiser decisions." Right? Not always—for LLMs, just like humans. We bring this scrutiny to AI by adapting the Hidden Profile paradigm from social psychology to assess how multi-agent LLMs actually reason…

Is there cryptography / HCI work on generating proofs of inability to comply, e.g. to prevent rubber hose attacks on non-custodial crypto wallets? Something like blind setup with social recovery paths (t of n) unknown to you, with time-locking + dead-man's switches for coercion.

🚨 Introducing our @TmlrOrg paper “Unlearning Sensitive Information in Multimodal LLMs: Benchmark and Attack-Defense Evaluation” W:nt UnLOK-VQA, a benchmark to evaluate unlearning in vision-and-language models—where both images and text may encode sensitive or private…

wrote this a month ago & 15 people liked it & that was enough to make me happy & put it out of mind. this weekend someone at substack found it & featured it & I went from around 100 subscribers to almost 4k over a few days. surreal.

Introducing @operative_sh MCP - 📍Loads up a @browser_use agent on your web app - 🖱️Click-tests your app like a human would - 🖥️Collects the console / network logs, errors and screenshots to return in MCP response - Now w/ 600+ Github🌟s: Let the coding agent debug itself.

Overthinking === Underfeeling

Now introducing UI vibe-testing with @operative_sh! We made an MCP that - 🤖 Launches a @browser_use web agent to go vibe-test your web app -⚡️ Makes browser-use 2x faster with Gemini 2.0 -🪄 Debugs with console + network logs Your coding agents can now self-debug without you!

Managing my emotional health has been something no one prepared me for in my research career and life as a young adult. I wrote this article to share some wisdom I've gained over the past few years. pradyumnashome.medium.com/the-emotional-…