Tag
A prediction suggests that continual learning will become adoption-ready for local AI systems by 2027, criticizing current methods like markdown harnesses and in-context learning.
This paper proposes Homeostatic Continual Learning, a method that enables AI agents to learn continuously in changing environments without catastrophic forgetting by detecting outliers and incrementally expanding world models.
ScienceBuddy introduces a recursive-in-recursive self-improvement paradigm for interactive scientific agents, enabling continual evolution through researcher collaboration and feedback.
This paper introduces ASCIL, a post-ASR framework that adapts to user feedback and contextual signals to correct false wake-up activations in AI assistants, achieving significant error reduction with low latency.
This paper proposes an audit protocol for update admission in continual embodied agents, emphasizing the need to balance error control with retained learning opportunities to prevent harmful updates and enable useful learning.
The paper introduces a fork ledger protocol to measure the counterfactual utility of world-model updates in continual adaptation, showing that always applying updates can lower performance on control tasks like CartPole and Walker.
This paper presents a Bellman optimality equation for optimizing plasticity in continual reinforcement learning, framing the stability-plasticity tradeoff as an empowerment-plasticity tradeoff.
AhaBench is a benchmark suite that evaluates whether language agents improve from prior experience in long-horizon tasks by testing exploration, knowledge transfer, and delayed feedback handling.
Fly Escape Room is introduced as the world's first game with NPCs powered by a real fruit fly brain, simulating olfactory intelligence and providing a platform for continual learning experiments.
Supermemory highlights the importance of continual learning for AI agents and introduces learner-1, a new model designed to advance memory and in-context learning for various use cases.
This paper shows that combining complementary continual learning mechanisms enhances long-horizon memorization in language models, boosting retention by 28-fold through data, function, and weight anchors with merged LoRA.
The article speculates that GPT-6 Astra Aeon, an AI agent with persistent memory, could enable continual learning and represent a significant shift in AI capabilities beyond benchmark improvements.
This paper presents a theoretical framework showing that learning in neural networks generates graph symmetries (fibrations and coverings), which enable drastic model compression and improve continual learning performance.
SkillGLoW is a method for LLM agents that organizes skills into procedural families to enhance self-improvement on long-horizon tasks, demonstrating significant performance gains over baselines.
MASkills presents a continual learning framework that optimizes multi-agent LLM systems through agent skills, using skill-conditioned credit assignment and hierarchical aggregation to improve performance on tasks like HotpotQA and GAIA.
The paper introduces EVOHARNESSBENCH, a benchmark for evaluating LLM agents under evolving tool, skill, and agent harnesses, revealing gaps in retention and adaptation.
The article discusses Test Time Training as a potential new scaling axis in AI development, analyzing a paper that frames it as a form of linear attention and exploring its implications for model training and continual learning.
LangChain is hosting an evening event in San Francisco featuring technical talks on continual learning, speakers from Prime Intellect and Baseten, and networking opportunities for the AI agent engineering community.
HypReflect is a framework for continual personalization in LLM assistants that infers explicit preference hypotheses from user signals and uses self-distillation to adapt, outperforming baselines and generalizing to unseen users and cross-domain settings.
The article proposes a long-term experiment to preserve the developmental continuity of an embodied AI agent over decades to study consciousness and individuality, comparing it with newly initialized agents.