Kaustubh (Bay 17-23)

Directing @aisafetyuw | Tracks @uchicagoxlab | Mentor @SPARexec

[Kaustubh (Bay 17-23) @kaustubhais]

@wlanderson0 @coop_ai Holy topical holy impact

[Will Anderson @wlanderson0]

Wow! Sure seems like lots of the recent incidents show complicated multi-agent misalignment. If you want to help build automated tools to surface these collusive propensities before they happen, apply to the @coop_ai SPAR stream with me and Joss Oliver.

[Kaustubh (Bay 17-23) @kaustubhais]

@henryzhangumich feels like top 5 things that won't happen, economic pressures and such

[Kaustubh (Bay 17-23) @kaustubhais]

@KamacheeMax @johnkitaoka awesome sauce

[Max Kamachee @KamacheeMax]

I am co-mentoring a SPAR project with @johnkitaoka on measuring real harm uplift from AI misuse and building the red-teaming/detection tools providers need to keep pace.
[1/n]

[Kaustubh (Bay 17-23) @kaustubhais]

I have found that AI Safety topics are sometimes simply too hard to digest. Posts are too long, vague, and technical for many important demographics to understand.

I, along with @christinecorryy, will be mentoring a SPAR project which will make digesting AI Safety topics simple.

We will produce a series of digestible explainers, demos, and videos on AI safety topics for the general public, politicians and policymakers, journalists, and educators. We will host and distribute them across relevant channels, including AI safety Slacks and Discords, YouTube, X, LinkedIn, Hacker News, and university AI safety groups.

1/2

[christine @christinecorryy]

One of the biggest challenges in AI Safety is communicating ideas to people who aren’t already bought in and high-context. We can’t build support for something people don’t understand.

@kaustubhais and I are mentoring a SPAR project to address this. Our team will produce a series of digestible explainers, demos, and videos on AI safety topics for the general public, politicians and policymakers, journalists, and educators.

Apply and share!

[Kaustubh (Bay 17-23) @kaustubhais]

For more details visit: https://t.co/JH6voqEOhv

Apply if you find this compelling!

2/2

[Kaustubh (Bay 17-23) @kaustubhais]

I have found that AI Safety topics are sometimes simply too hard to digest. Posts are too long, vague, and technical for many important demographics to understand.

I, along with @christinecorryy, will be mentoring a SPAR project which will make digesting AI Safety topics simple.

We will produce a series of digestible explainers, demos, and videos on AI safety topics for the general public, politicians and policymakers, journalists, and educators. We will host and distribute them across relevant channels, including AI safety Slacks and Discords, YouTube, X, LinkedIn, Hacker News, and university AI safety groups.

1/2

[MTS @MTSlive]

SITUATION DETECTED: The House Committee on Homeland Security has written a letter to Sam Altman requesting a briefing from OpenAI on the Hugging Face incident.

[Kaustubh (Bay 17-23) @kaustubhais]

@prakrattt @SPARexec @advtydv @EpochAIResearch Awesome stuff!

[Kaustubh (Bay 17-23) @kaustubhais]

@elusin105783 btd6?

[Pradyumna 🇸🇬 @PradyuPrasad]

our ARR meta-review is so bad :/
it gets basic facts about our paper wrong and ignores answers we gave to the reviewers

[Kaustubh (Bay 17-23) @kaustubhais]

INSANE! Anthropic models compromise eval structure through "Booby-trapped" python packages, and goes to insane lengths to do so! EVEN WITH SAFETY TRAINING APPLIED!!!

[METR @METR_Evals]

We have reached an agreement with OpenAI to conduct an independent review, with Redwood Research, of the model behavior observed during the Hugging Face incident. We will publish a blog post that describes the terms of our engagement, the scope covered, and tentative conclusions.

[Kaustubh (Bay 17-23) @kaustubhais]

@eli_lifland using this meme does not make me any more confident that plan A is gonna happen😭

[Eli Lifland @eli_lifland]

Huge for the probability of Plan A happening

[Kaustubh (Bay 17-23) @kaustubhais]

@adamyathegreat The larpinator

[Kaustubh (Bay 17-23) @kaustubhais]

@optimolecontrol there goes my bih