Berkeley Team Teaching AI Language Like a Baby Wins Schmidt Sciences Funding
begusgasper · x · 2026-10-20
UC Berkeley linguistics professors Gašper Beguš, Nina Beguš, and colleagues are exploring how AI models can learn language the way infants do—by imitating speech rather than relying on massive text corpora. Compared with the 570GB of text (roughly 225 billion words) used to train GPT-3, children acquire language far more efficiently and with far lower energy costs.
More from Research
- SwitchSplat brings switchable object states to 3D Gaussian Splatting, headed to SIGGRAPH Asia 2026 — taiyasaki · 2026-10-20
- EgoXperience turns egocentric video into world-grounded 3D reconstructions in one forward pass — liuziwei7 · 2026-10-20
- Fudan & Shanghai AI Lab map PEFT recipes for MoE LLMs: tune experts alone, LoRA/DoRA win — kalyan_kpl · 2026-10-20
- Pramāṇa launches: 105k texts, 414M words of Indian primary sources, free to search — CatAstro_Piyush · 2026-10-20
- RoboCloth: 500 real cloths, 80B observations to replace 2003-era BRDF data — taiyasaki · 2026-10-20
- NVIDIA OmniDreams: real-time generative world model for closed-loop AV simulation, at 16 GB300 per video — yassineyousfi_ · 2026-10-20