New paper lifts LLM memory retention 28x to 34.9% across 100 sequential tasks
DanielKhashabi · x · 2026-09-16
A new paper composes continual learning mechanisms to raise final retention on 100 sequential tasks from 1.2% to 34.9% — a roughly 28x improvement. The associated dataset has been released on Hugging Face for reproduction and follow-up work.
More from Models
- SOTA GGUF quants for Qwen3.8-Flash-Next: near-baseline quality at 1/5 the size — BullfrogScary8947 · 2026-09-16
- Users report Gemini Flash seems to have lost all memory of conversations — Welvex · 2026-09-16
- ChatGPT was the fastest-growing top-10 website in August as students returned to school — menhguin · 2026-09-16
- Qwen 27B on a single RTX 5090 generates 8-scene animation from one prompt — Acceptable-Object390 · 2026-09-16
- ChatGPT Plus users report image generation quality appears degraded — nilanganray · 2026-09-16
- Claude usage bug unexpectedly grants extra quota as user plans ASC CLI release — rudrank · 2026-09-16