Stanford, Berkeley introduce TokenGate attention method

Researchers from Stanford and UC Berkeley introduce TokenGate, a token-gated attention mechanism that makes long-context prefill up to 11.3x faster, potentially reducing compute costs for long-text inference.

• AI Use Leads Berkeley Professor to Stop Grading Homework2026-10-25
• Stanford et al. unveil Hybrid Reasoning Agents2026-10-16
• Study: 10 Minutes of AI Use Erodes Persistence2026-10-11
• Sisters launch AI dating advice startup Hot Girl Hotline2026-10-05
• AI system Ataraxos beats top human Stratego player2026-10-01