Apart Research - Digital Minds Research Sprint
Digital Minds Research Sprint — Saarbrücken Site
Hosted by AI Safety Saarland · Local hub for the Apart Research sprint
Please register on the Apart website before you RSVP here. Apart webpage for the Digital Minds Research Sprint: https://apartresearch.com/sprints/digital-minds-research-sprint-2026-08-14-to-2026-08-16
Frontier models express values, report internal states, and act as though they have interests — yet we have no reliable way to tell a genuine preference from a portrayed character. Over one weekend, we'll design and run the experiments that build the empirical foundations of AI welfare.
This is the Saarbrücken in-person hub for the Digital Minds Research Sprint, a global 3-day research sprint by Apart Research, co-organized with the NYU Center for Mind, Ethics & Policy and Eleos AI Research. Join us to work through the weekend alongside other participants in Saarland instead of alone at home — the wider sprint runs online with hubs in San Francisco and Berlin.
When: Friday, 14 August – Sunday, 16 August 2026
Submissions close Sunday 16 August, 11:59 PM Anywhere on Earth.
Where: details to be sent to registered participants.
What you'll do
You'll form a team and produce a short research report (optionally with code and a demo) probing the preferences, welfare signals, introspective abilities, and identity of frontier models. Six tracks to anchor your project:
Model Preferences & Trade-offs — elicit pairwise preferences across reframings; measure transitivity, coherence, and stated vs. revealed preferences.
Distress, Flourishing & Valence Signals — map the contexts that correlate with negative or positive-valence outputs; test whether the signals are stable or surface artifacts.
Introspection & Self-Report Reliability — replicate concept-injection tests, compare naive vs. structured elicitation, probe privileged access.
Preference Elicitation Methods — build and validate the measurement tooling the other tracks depend on; test multi-method convergence.
The Assistant Persona & Model Identity — is the assistant a model, an instance, a persona? How stable is it, and can it mask underlying preferences?
Open / Novel Considerations — bring your own angle.
Prizes
$2,000+ in cash prizes (breakdown announced before the sprint). The winning team is invited to ConCon, the Eleos AI Research conference on AI consciousness and welfare (18–20 September 2026, Lighthaven, Berkeley). Top teams are invited to apply to the Apart Fellowship, a 3–6 month research accelerator with mentorship, funding, and publication support. Mentor introductions and publication support for winners; selected projects may be shared on the Alignment Forum and LessWrong.
No prior background in AI safety or AI welfare is required. All you need is curiosity and a willingness to scope a tight empirical question. Bring a laptop; a curated reading list is available on the sprint's Resources tab.
Important: register for the sprint itself on the Apart Research page in addition to signing up here, so you're in the team-matching and submission system. This page is for the local Saarbrücken site only.
Questions about the venue and local logistics: reach out to AI Safety Saarland.
Questions about the sprint: sprints@apartresearch.com · Join Discord Server