HERD: Harness-Evolving Recursive Distillation for Continual Learning
Frames the problem: parametric forgetting versus non-parametric limits, with a distillation approach.
AI Storyline
3 items · 2 sources · 3 days
Operational story trace
Continual Learning
Follow in this browser to see new updates on your Live feed.
Current stateDeveloping · research
Latest change
CAMR (Oct 27) proposes compatibility-aware memory replay, deciding whether incoming data is worth replaying before it is used to update an LLM.
Earlier contextThe story so far
An Oct 22 paper (HERD) framed continual learning as a trade-off between parametric methods, which forget, and non-parametric ones. Three days later a benchmark began measuring whether LLM agents actually improve from experience over long horizons.
editor-curated · source-linked
Arc
Oct 22Oct 27 · now
FRAMING · Oct 22HERD targets forgetting with recursive distillation1 source · show source ▾
MEASURE · Oct 25Continual Learning Exam benchmarks LLM agents over long horizons1 source · show source ▾
Continual Learning Exam: Benchmarking Long-Horizon RSI and Experiential Learning in LLM Agents
Adds a benchmark for agents learning from experience over long horizons.
NOW · Oct 27CAMR selects what to replay by compatibility1 source · show source ▾
CAMR: Compatibility-Aware Memory Replay for Continual Learning in Large Language Models
Adds a replay-selection method that judges whether new data is compatible before use.
Day 1 Thursday, Oct 22, 2026
HERD: Harness-Evolving Recursive Distillation for Continual Learning
Frames the problem: parametric forgetting versus non-parametric limits, with a distillation approach.
Day 2 Sunday, Oct 25, 2026
Continual Learning Exam: Benchmarking Long-Horizon RSI and Experiential Learning in LLM Agents
Adds a benchmark for agents learning from experience over long horizons.
Day 3 Tuesday, Oct 27, 2026
CAMR: Compatibility-Aware Memory Replay for Continual Learning in Large Language Models
Adds a replay-selection method that judges whether new data is compatible before use.
What to watch — open questions
- Do any labs report results on the Continual Learning Exam benchmark?
- Do HERD or CAMR release code or reproducible numbers on production-scale models?
How this thread was built
editor wrote the arc · 3 beatswatcher 1 status change
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.