Berkeley's Beguš uses AI interpretability to find vowel analogs in whale communication
begusgasper · x · 2026-11-04
Cambridge NLIP online seminar: UC Berkeley's Gašper Beguš presents a framework modeling language as informative imitation. He introduces ciwaGAN, which learns from raw unsupervised audio with communicative intent and embodied representations, plus the CDEV interpretability technique enabling linguistic and neural simulations. Case study: CDEV uncovered analogs of human vowels in whale communication, showing how understanding generative models can drive scientific discovery.
More from Research
- STOC submissions include dueling P=NP and P!=NP proofs; blogger bets both wrong — thegautamkamath · 2026-11-04
- DeepSeek V4.1 Flash reads raw weights: 3 of 4 RL checkpoints identified correctly — poppear · 2026-11-04
- CoworkOS Debuts: Three-Layer Agent Coworker Framework Lifts Qwen3.6 by 9.76 Points — KyeGomezB · 2026-11-04
- STOC 2027 draws 1665 submissions and publishes all of them, with author names, immediately — thegautamkamath · 2026-11-04
- DOLORES: Training-Free Meta-Scaffold Lifts 8B Models Past 32B Baselines — StellaLisy · 2026-11-04
- Formalizing PTX in Lean to formally verify GPU kernels written by AI agents — gharik · 2026-11-04