## Conferences before  ## Conferences before  ## Solution: ### Workshop on bitter lessons! - Three sessions with three speakers talking about the bitter lesson (and more broadly the AI "explosion") + panel discussions - Jon Barron: "3D = bitter lessonned?" - Vincent Sitzmann: "3D (and more broadly mid-level vision) is useless" - Bharath Hahirahan: "Mid level vision is useful" ## Is 3D useful? Two contradictory signals: - Workshop on bitter lessons: avoid 3D when you don't need it *explicitely* - Last 3 best papers: - Dust3r (unofficial) - VGGT - D4RT - Small scale experiments on transfer learning for VGG-$\Omega$ (+ numerous papers on improving video generation using explicit 3D). ## Random takes {.smaller} - 3D for vision might be dead, but vision for 3D is definitely alive [@Li2026ART; @Liu2026DeepFeature; @Xiao2026Universal3D]. - "3D is dead" ≠ "Geometry is dead" [@Chiang2026COTFM; @Luo2026FlowMatching] - 2026: LLMs still don't improve the quality of orals or posters - TRELLIS structured latent space enables numerous applications (tokenization[@dutt2026lost], editing[@Hu2026Easy3E], morphing[@Sun2026MorphAny3D], improving trellis[@Xia2026Pointsto] (again), ...) - *Weak* signal that DinoV3 is not that better that DinoV2 ## BrickNet: Graph-Backed Generative Brick Assembly  ## BrickNet: Graph-Backed Generative Brick Assembly  ## Can You Learn to See Without Images? Procedural Warm-Up for Vision Transformers  ## Can You Learn to See Without Images? Procedural Warm-Up for Vision Transformers  
