Vishesh

Senior Software Engineer @Google AI, CAIR, ex SDE 2 @Amazon, Sharechat

[Vishesh @vishessh_t]

https://t.co/FxLpCGfu2C

[Vishesh @vishessh_t]

https://t.co/mOOj3sHcxq

[Vishesh @vishessh_t]

📣📣Excited to share our new paper: ASPIRE. In this paper we share a new framework for discovering vulnerabilities in LLM agents.
ASPIRE acts like a security researcher. It builds a "Behavior Graph" to map out the agent's logic, then uses "Explore" and "Exploit" experts to systematically hunt down hidden vulnerabilities across the agent's attack surface.
It’s highly effective: on recent benchmarks, ASPIRE uncovered deep, multi-step exploits with up to a 54% attack success rate, costing less than $2 per discovered vulnerability.
If you are building or securing autonomous agents, you can read the full methodology in https://t.co/GBCAIeSQ6B

Thank you to all the authors/coauthors. #AgentSecurity, #AgentSafety, #AgentQuality

[Vishesh @vishessh_t]

📣📣Excited to share our new paper: ASPIRE. In this paper we share a new framework for discovering vulnerabilities in LLM agents.
ASPIRE acts like a security researcher. It builds a "Behavior Graph" to map out the agent's logic, then uses "Explore" and "Exploit" experts to systematically hunt down hidden vulnerabilities across the agent's attack surface.
It’s highly effective: on recent benchmarks, ASPIRE uncovered deep, multi-step exploits with up to a 54% attack success rate, costing less than $2 per discovered vulnerability.
If you are building or securing autonomous agents, you can read the full methodology in https://t.co/GBCAIeSQ6B

Thank you to all the authors/coauthors. #AgentSecurity, #AgentSafety, #AgentQuality

[Nithish Kannen @NithishKannen]

This Flash series beating models of much bigger size and scale is a result of general post-training breakthroughs internally. The future looks exciting!

[CG @cgtwts]

Moonshot AI CEO Yang Zhilin just gave one of the clearest masterclasses on how frontier AI models are actually built.

[Vishesh @vishessh_t]

Good read. Talks about creativity of diffusion models ( its ability to generate novel data rather than memorize) https://t.co/fPj5WUhix8

[Phil Chen @philhchen]

https://t.co/oMoEAYkqGl

[Nithish Kannen @NithishKannen]

@cneuralnetwork https://t.co/wSlenAAR0L

[Vishesh @vishessh_t]

Good read https://t.co/rBE8ent7ix

[Moysei @0xMoysei]

Karpathy just wrote the manual for Claude + Obsidian as a real second brain.

Most vaults die the same way. A year of saved articles and highlights. None of it linked. The graph rots while it still looks impressive.

So he moved the upkeep to the model. You curate sources and ask questions. Claude files, links, and reconciles. You keep judgment. It keeps the books.

raw belongs to you and never gets edited. wiki belongs to Claude. It isn't RAG. Your sources compile once into linked pages and compound from there.

9 rules. Start with 10 sources, not 10,000.
Most people hoard notes. This turns them into a brain that maintains itself.

[Raytar @Raytar]

Andrej Karpathy joined Anthropic five weeks ago.

Yesterday my friend on his team sent me the Claude.md file he actually uses.

It completely changed how I work with Claude.
From the very first message, the difference was obvious.

With this file, Claude finally stops fighting me and starts working exactly the way I need it to.

Bookmark it before it gets taken down.

Read it now, then check the article below.

[Vishesh @vishessh_t]

https://t.co/FxLpCGfu2C

[Vishesh @vishessh_t]

📢📢 Excited to share our new paper on LLM Agents safety. "PI-Hunter: Automated Red-Teaming for Exposing and Localizing Prompt Injections in LLM Agents!"

[Chrome @0xchromium]

Andrej Karpathy spent 2h showing how he actually uses AI day to day

he's a co-founder of OpenAI and led AI at Tesla, so when he shows how he works, it’s worth watching

and the whole session is just him telling the machine what he wants in simple terms, like he's briefing a coworker

watch what's actually happening the entire time:

> he describes the task in normal words
> it goes off and does the work
> he glances at the result and nudges it with one more sentence

that's the whole skill, and you've had it since you learned to talk

the only gap between that and a worker that runs on its own is handing that sentence a schedule and the tools to act

check his work, then build the version that keeps working when you stop

[Tom Yeh @ProfTomYeh]

Here's my beginner's lecture series for RAG, Vector Database, Agent, and Multi-Agents: Download slides: 👇
* RAG: https://t.co/O3OvTXVWUI
* Agents: https://t.co/BeJedAKwC9
* Vector Database: https://t.co/UZw5Lieszp
* Multi-Agents: https://t.co/iEMJKeMYAE

---
100% original, made by hand ✍️
Join 56K+ readers of my newsletter: https://t.co/fFt8roc8D9

[Tech with Mak @techNmak]

Last month, Google dropped something interesting:
five AI Agent papers released across five consecutive days, one per day, each digging into a different part of how agents should be built, evaluated, secured, and deployed.

No big splash, just a steady rollout of more than 250 pages of highly technical material.

Here’s the distilled version of what those five drops covered:

Read. Learn. Bookmark.

[Nithish Kannen @NithishKannen]

Nano Banana Pro planned @Roprajo's Cocktail Menu yesterday!

[Susan Zhang @suchenzang]

🤔🤔🤔

[Vishesh @vishessh_t]

@NithishKannen @GoogleDeepMind Congratulations 🎉. Well deserved 👏👏. See you around 😀