New Research: Generative Models Can Insert Tokens Anywhere
thjashin · x · 2026-07-06
Traditional autoregressive (AR) models are limited in expressiveness as they can only append tokens to the end of a sequence. This paper introduces "insertion-based generative models" that allow token insertion at arbitrary positions, offering stronger theoretical expressiveness but posing challenges for standard maximum likelihood training.
To address this, the research introduces a scalable variational learning method, providing a novel approach for generative models looking to move beyond the AR paradigm.
More from Research
- Kimi K3 may be strong on cyber, but token efficiency keeps it off UK AISIS — teortaxesTex · 2026-07-27
- ARC AGI 3 should have stayed private, with no examples or public dataset — flowersslop · 2026-07-27
- ExploitGym may have only 60–70% solvable tasks, fueling the OpenAI cheating debate — max_paperclips · 2026-07-27
- RTX 5090 local tests show Qwen Q6 can drop to 15 tok/s at 80k context — LFAdvice7984 · 2026-07-27
- Noahpinion quotes Chollet: intelligence may hit a hard ceiling — binarybits · 2026-07-27
- Paper argues graph topology can become the core operating system for AI agents — theomitsa · 2026-07-27