Déjà View: a looped 117M model rivals billion-parameter 3D reconstruction

Shows that looping one transformer block can replace much of the depth in feed-forward 3D reconstruction models. It also exposes iteration count as an inference-time compute knob, a trade-off practitioners can reuse.

• transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.
• inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
• test-time compute — Spending more computation when the model answers — thinking longer, trying multiple attempts — to buy accuracy without training a bigger model.

• articleVerifier-driven tuning took a chip design agent from 55% to 95%
• articleNVIDIA releases Nemotron 3.5 Bolt, a 4GB open model for local agents
• articleStepFun Step 5 Preview: 600B open-weight agent model with NVFP4 recipes
• articleNVIDIA's PivotOPD trains agents to recover from early mistakes

• articleParcae: Doing more with fewer parameters using stable looped modelsTogether AI
• articleTencent open-sources WorldMirror for single-pass 3D reconstructionTencent Hy
• postThree World Labs papers turn 2D generative priors into 3D reasoningWorld Labs

Checking sign-in…

Loading comments…