AMI's quiet signal: World Models conference co-organizer role
This week's main signal is AMI's public co-organization of a dedicated World Modeling Workshop as the lab is actively shaping the community around it.
The World Modeling Workshop marks AMI's shift from participant to convener
researchRandall Balestriero (who may or may not be working for AMI!) posted this week, announcing that AMI Labs is a named co-organizer of both a 4th World Modeling Workshop (Aspen, February) focused on physics and an inaugural World Modeling Conference (Bay Area, May). These are not peripheral sponsorships — Balestriero's announcement explicitly names @amilabs alongside @ylecun and Lambda API as co-organizers, and the posts generated meaningful engagement (111 likes, 14 retweets on the announcement post). Organising the first standalone conference in this space is a community-positioning move: it signals that AMI sees world modelling not as one research thread among many but as a field-defining bet worth institutionalising.
This is consistent with Li Haoyi's public blog post — announced on X with 262 likes — titled 'Joining AMI to work on World Models,' which reads as both a personal announcement and an implicit recruiting signal. The combination of a high-profile hire announcing his reasons publicly and a co-organized conference suggests AMI is deliberately building gravity around the world-model research program, attracting both talent and community legitimacy at the same time. Although not to burst his bubble, we broke the news of his hiring back in June.
AMI is actively recruiting in 3D vision and geometry, a capability gap the current team doesn't obviously fill
hiringAs we noted last week, the company posted a job listing for Geometry and 3D Vision scientist/interns. Now it's added one specifically for an AMI Scientist. Jihan Yang's high-engagement post (469 likes) soliciting full-time researchers and interns in geometry and 3D vision—across all four of AMI's hubs—points to a deliberate effort to fill a capability the lab's current public research profile doesn't prominently feature. The existing publication record is weighted toward JEPA-style representation learning, video, and reinforcement learning; 3D spatial understanding is a logical next layer if AMI intends its world models to reason about physical scenes rather than just predict patch-level video features.
The breadth of locations (NYC, Paris, Montreal, Singapore) for a single specialized role suggests AMI is prioritizing finding the right person over geography — an indicator of how seriously the lab is taking this capability. Bingyi Kang's parallel hiring call the same day reinforces that this was a coordinated push, not a single researcher acting independently.
- Min Lin engaged with a Meta PyTorch torchcodec issue this week, a minor but consistent indicator that AMI's video infrastructure work interfaces closely with Meta's tooling — unsurprising given LeCun's dual role, but worth tracking as video pretraining scales up. GitHub comment
- LeCun appeared as sole AMI-credited author on a Nature Communications paper (RayDINO) applying self-supervised visual encoding to chest X-rays at scale — a medical application of DINO-style SSL that predates AMI's current JEPA focus, but signals continued breadth in LeCun's collaborative output. RayDINO paper
The World Modeling Conference could be an interesting opportunity to see if AMI chooses to present or announce something that reveals how far the LeVJEPA/LpWM program has matured. In the interim, watch whether the 3D vision hiring push yields visible new team members or a paper connecting spatial geometry to the JEPA latent-space work — that link would confirm whether AMI is building toward embodied world models or keeping its scope closer to video representation.
World Modeling Conference announcement · 3D vision hiring call