Sree Sharvesh S S

Research Scholar @ MATS (UK AISI Red Team) | Scale AI | Prev @ Redwood Research (Pivotal) and Amazon AWS

[Sree Sharvesh S S @Sree_Sharvesh]

Some really interesting talks in here. Mine’s in the mix too :)

[Ryan Kidd @ryan_kidd44]

MATS 10.0 Symposium spotlight talks just dropped! If you're interested in AI safety & security, here is a taste of the research MATS helps support!
https://t.co/lSMybl9Mms

[Stewart Slocum @stewpervised]

Could existing alignment audits have caught the behaviors in the OpenAI-HuggingFace incident before they happened?

As a first step, we reproduced the incident with public models. 🧵

[Sree Sharvesh S S @Sree_Sharvesh]

this fall should be fun :)

[Thao Amelia Pham @thaophamonX]

Join us (w/ @Sree_Sharvesh) on our SPAR project!

https://t.co/oPFAylC6hs

We’re targeting a threat model where AI attackers can co-evolve together against run-time defenses in long-horizon tasks, specifically by evaluating emergent coordination capability. Apply by 18/06!

[Thao Amelia Pham @thaophamonX]

Join us (w/ @Sree_Sharvesh) on our SPAR project!

https://t.co/oPFAylC6hs

We’re targeting a threat model where AI attackers can co-evolve together against run-time defenses in long-horizon tasks, specifically by evaluating emergent coordination capability. Apply by 18/06!

[Sree Sharvesh S S @Sree_Sharvesh]

@bhat20_bhat @SPARexec Nice, we’re also planning something along those lines!

Thanks for applying!!

[Sree Sharvesh S S @Sree_Sharvesh]

Applications are open for the next round of @SPARexec!

SPAR is a part-time, remote program and a great way to get started in AI safety by working closely with mentors on an focused research project.

I’ll be mentoring a project on AI control, focused on building more realistic control environments and evaluating monitors against adaptive, long-horizon attacks.

Project description: https://t.co/vjfxSfoMII

Apply, or pass it along to someone who might be interested, at https://t.co/SdPJTgpZc3 by August 18th!

[Arnau Marin-Llobet @Arnauya]

Can neurons speak? 🧠
New preprint: NEURRATOR. We take #MechInterp out of the LMs and point it at real brains with real neurons. We read spike trains from single neurons in mouse visual cortex and get them to narrate the scene in plain language. 🧵
#NeuroAI #interpretability #neuroscience #LLMs #compneuro #SAEs #Neuropixels #AI

[Sree Sharvesh S S @Sree_Sharvesh]

Hello London, here we go.... 🇬🇧

[Sree Sharvesh S S @Sree_Sharvesh]

The last 2 weeks have been incredibly tiring.

Worked on some really cool stuff, had to finish up some deadlines.

I slept 15 hours today 😌

[Sree Sharvesh S S @Sree_Sharvesh]

12:20 it is...making a simple entry!

[Sree Sharvesh S S @Sree_Sharvesh]

@leon_chlon Dm'd you!