Arkose is an AI safety fieldbuilding organisation that supports experienced machine learning professionals — such as professors and research engineers — to engage with the field. We focus on those new to AI safety, and have strong evidence that our work helps them take meaningful first steps.

Since December 2023, we’ve held nearly 300 one-on-one calls with mid-career machine learning researchers and engineers. In follow-up surveys, 79% reported that the call accelerated their involvement in AI safety^[1]. Nonetheless, we’re at serious risk of shutting down in the coming weeks due to a lack of funding. Several funders have told us that we’re close to meeting their bar, but not quite there, leaving us in a precarious position. Without immediate support, we won’t be able to continue this work.

If you're interested in supporting Arkose, or would like to learn more, please reach out here or email victoria@arkose.org. You can also make donations through Manifund, which will be refunded if we do not meet our minimum bar.

What evidence is there that Arkose is impactful?

AI safety remains significantly talent-constrained, particularly with regard to researchers who possess both strong machine learning credentials and a deep understanding of existential risk from advanced AI.

Arkose aims to address this gap by identifying talented researchers (e.g., those with publications at NeurIPS, ICML, and ICLR), engaging them through one-on-one calls, and supporting their immediate next steps into the field.

Following each call, we distribute a feedback form to participants. 52% of professionals complete the survey, and of those, 79% report that the call accelerated their involvement in AI safety:

We also send a short form 6 months later, which demonstrates that professionals take a variety of concrete actions following their call with us:

We think that accelerating a senior machine learning professional's involvement in AI safety (e.g. a professor) by 1-3 months is a great outcome, and we're excited about the opportunity to improve our call quality over the next year and support researchers even better.

Unfortunately, we're not able to share all of our impact analysis publicly. If you're interested in learning more (for instance, if you'd like to see summaries of particularly high or low impact calls), please reach out to team@arkose.org.

What would the funding allow you to achieve?

We nee...

Posts

Wikitag Contributions

Comments