Welcome to the Apart Updates, your portal to the Apart community's research and events! Subscribe to hear the latest from us every three months.

Our first issue of the Apart Updates is a quick read. Highlights include:

• 🥂 Celebrating the Apart community and AI security at NeurIPS

• 👩‍🔬 Entering 2024: Kicking off a new Lab cohort, preparing our Sprints schedule, and introducing Apart Updates

  • 🏛️ Join the AI Governance Sprint in just 6 days!

  • 🙋‍♀️ Support our continued work at Apart

We are thankful for all of your engagement in 2023. Happy New Year from all of us at Apart!

Wrapping up 2023

Apart closed off a great 2023 by joining 15,000 other AI researchers at the NeurIPS conference in New Orleans. It was motivating to see four papers at NeurIPS from projects created at our Apart Sprints. We were there ourselves to present two papers from Apart Lab and participate in an insightful panel debate about AI and security. 

Significantly more safety-related papers came out compared to last year but not many compared to the 3,584 total accepted papers. It motivates us to see that there's still a big gap to get more technical research towards the question we consider important: how can we ensure that AI develops responsibly while not posing societal-scale risks?

As we approach the end of an eventful year at Apart, we want to extend our heartfelt thanks to everyone in our community for their active involvement in this critical discourse. Let's reflect on the year that was and look forward to more groundbreaking work ahead!

Celebrating the Apart community

In the past year, we hosted 17 research sprints with more than 1,000 participants! 🥳

Overall, we have seen an impressive 170+ resulting projects from our Alignment Jams, with 10 out of our 24 Apart Lab fellows already completing their research, resulting in three publications accepted at top-tier academic ML venues and workshops at NeurIPS, ACL & ICLR – big thank you and congratulations to Julia, Albert, Alex, Jason, and the many researchers involved :) 

Check out other recent community milestones:

• 4 papers from the Sprints were present at the SOLAR, MASEC and XAI in Action workshops at NeurIPS! One as a Lab project and three others developed by the Apart community. Check out "Assessing Risks of Using Autonomous Language Models in Military and Diplomatic Planning", "Towards a Situational Awareness Benchmark for LLMs" (watch an Apart talk by one of the authors here), "Second-order Jailbreaks: Generative Agents Successfully Manipulate Through an Intermediary" and "DeepDecipher: Accessing and Investigating Neuron Activation in Large Language Models" (check out the tool here).

• 21 projects were submitted to the Multi-Agent Safety Hackathon co-organized with the Cooperative AI Foundation in October; some of these are going into a large report on multi-agent risk from advanced AI that we are looking forward to reading in 2024.

• 2 more papers by Lab associates were published at ACL and the RTML workshop at ICLR; check out "Detecting Edit Failures In Large Language Models: An Improved Specificity Benchmark" and "Neuron to Graph: Interpreting Language Model Neurons at Scale" from previous Lab members Julia Persson, Jason Hoelscher-Obermaier, and Alex Foote (among others)!

• 30 researchers participated in our in-person AI safety entrepreneurship hackathon hosted with EntrepreneurFirst in the Netherlands which was a fascinating experiment in making AI safety theory into reality; and we are sure to hear more from at least one of the teams.

• In 2023, we also shared our perspectives in talks and debates on AI safety and security at EAGx Nordics, Warsaw, Cambridge, and Virtual; Vision Weekend, the MASEC workshop, the Intelligent Cooperation Workshop, at the Apart Sprints (watch our recordings), and many other places. It motivates us a lot to see the engagement from the research community.

We also have great reasons to continue our excitement into January, with the addition of more core team members to Apart and seven more of our papers currently under review!

Our goal at Apart is to reduce global risks from AI by incubating novel AI safety research and enabling the next generation of global AI safety researchers. We have a firm belief that this goal is a community-wide effort and would like to express our gratitude to everyone who made 2023 possible. We look forward to doubling down on our efforts with everyone in 2024!

To this end, as we aim to create even better research and host even greater events, we would love to hear your opinions about AI safety and how we can make it even more inclusive and exciting to be a part of the Apart community.

Looking forward to 2024

There's a lot to look forward to in 2024 — for example the Apart Updates! The Updates will be your ~monthly update from the Apart community and include helpful resources for your own work along with research spotlights. We're excited to bring community news to everyone.

Kicking off a new Apart Lab cohort

We’re really excited that more than a dozen researchers will join us for an upcoming Apart Lab cohort starting in the second week of the new year. These fellows will work on AI safety topics such as the safety of multilingual and multimodal foundation models, cross-lingual robustness of alignment techniques and safety evaluations along with the robustness of AI evaluations in general. We’re especially excited about this new cohort since we have made a lot of improvements to the fellowship, which we are eager to see in practice!

Are you excited about Apart?

In 2024, we will double down on our current efforts in research and experiment with more ways of doing research. If you are excited about what we do, want to help us or wish to collaborate, please share your interest below.

Happy new year!

Thanks for reading thus far. We wish you an amazing 2024 and hope we will continue to work together for a world with safer AI. If you enjoyed what Apart has been up to, please forward this e-mail to friends and colleagues or sign up for our newsletter here.