Tagged articles
Continual Learning
40 articles · Page 1 of 1
Controlled experiments comparing MIT's Astra weight-editing approach with a hypernetwork method reveal that while Astra fixes repetition loops, it fails to internalize knowledge, personas, or skills, unlike hypernetworks which succeed; the authors argue weight editing alone is insufficient for stable learning and propose a complementary agent-hypernetwork workflow.
AstraContinual LearningHypernetworks
0 likes · 15 min read

HKU researchers propose RaPO, a reinforcement fine-tuning method that adds a retention reward based on trajectory-level drift to prevent catastrophic forgetting in multimodal large language models during visual continual learning, achieving state-of-the-art results on five benchmarks with minimal forgetting.
Catastrophic ForgettingContinual LearningGRPO
0 likes · 16 min read

Researchers from Zhejiang University, ByteDance Seed, and Tsinghua University propose Cortex, a method that consolidates contextual experience into model parameters via prospective rehearsal and context distillation, enabling agents to retain skills and knowledge without external retrieval, validated across tool use, fact updating, and agent tasks with minimal parameter growth.
Continual LearningKnowledge EditingLLM Agents
0 likes · 39 min read

Six AI researchers from academia and industry discuss recursive self-improvement (RSI), debating its definition, distinguishing it from AI-for-AI, exploring technical paths like data generation and harness optimization, identifying bottlenecks in failure diagnosis and environment feedback, reviewing benchmarks such as PostTrainBench and RSI-Bench, and speculating on future human roles in an RSI-driven world.
AI research automationAI4AIContinual Learning
0 likes · 69 min read

CMU researchers propose a unified training paradigm where pre-training, mid-training, supervised fine-tuning, and reinforcement learning share a single maximum likelihood objective, using data mixing and scheduling instead of stage-specific algorithms, and show improved reasoning and continual learning on puzzles, math, and code benchmarks.
CMUContinual LearningMaxRL
0 likes · 9 min read

Tongyan Lingzhi builds a continuous self-improving personal AI system using a shared base model plus lightweight Personal LoRA parameters, driven by a Recursive Self-Improvement (RSI) loop that learns from real user interactions, supported by training-inference parallelism (TIP) and Paged LoRA for efficient scaling.
AI infrastructureContinual LearningLoRA
0 likes · 15 min read

Researchers from UIUC and Tsinghua startup Astraculum built a Social World Model using a 7B LLM with continual self-distillation on 390 days of prediction market data, achieving deployment-time learning that beats static frontier models like GPT-5.6 and Claude Opus 5.
Continual LearningLLMPrediction Markets
0 likes · 11 min read

The article analyzes how large language models can move beyond temporary context or retrieval tricks by using knowledge editing to modify internal parameters, forming distributed knowledge loops, confidence‑guided belief updates, and a self‑sustaining continual‑learning cycle, while also serving as a scientific tool for probing model mechanisms.
AI researchContinual LearningKnowledge Editing
0 likes · 14 min read

Octopus introduces a history‑free gradient orthogonalization (HiFGO) constraint and a two‑stage finetuning strategy that together eliminate catastrophic forgetting, achieve positive backward transfer, and set new SOTA results on the UCIT benchmark, improving Avg and Last metrics by 2.14% and 6.82% without extra inference cost.
Continual LearningUCIT benchmarkgradient orthogonalization
0 likes · 11 min read

As AI models and frameworks turn into off‑the‑shelf components, enterprises must convert their unique business knowledge—task standards, evaluations, harnesses, and continual learning loops—into proprietary intelligent assets that persist beyond each model upgrade, creating a sustainable competitive moat.
AI agentsContinual LearningPost-Training
0 likes · 28 min read

The article surveys five emerging approaches to LLM continual learning—external agent memory, context engineering, post‑training, pre‑training, and self‑modifying models—explaining how each tackles the core obstacle of catastrophic forgetting, citing benchmarks such as TRACE, ACE, and SDFT, and reflecting on Karpathy’s ten‑year timeline.
Agent MemoryCatastrophic ForgettingContinual Learning
0 likes · 17 min read

The article examines the limits of Transformer models, outlines their three fundamental shortcomings, and evaluates the Mobius architecture that decouples knowledge and reasoning, showing up to four‑fold inference speed gains, comparable accuracy, and improved compositional generalisation.
AI EfficiencyContinual LearningMobius
0 likes · 14 min read

This survey reviews continual self‑supervised learning (CSSL) for vision models, explaining why self‑supervised objectives better resist forgetting, outlining evaluation protocols, categorizing six CSSL method families, and discussing challenges such as unified benchmarks, large‑scale models, and multimodal drift.
CSSLContinual LearningRepresentation Learning
0 likes · 16 min read

Google researchers propose a ‘sleep’ stage for large language models, where after active use the model consolidates recent experiences through knowledge‑seeding and self‑generated ‘dreaming’ tasks, addressing catastrophic forgetting and enabling continual learning.
AI researchCatastrophic ForgettingContinual Learning
0 likes · 11 min read

The paper introduces CaRE, a scalable continual‑learning framework that leverages a bi‑level routing mixture‑of‑experts to successfully train Vision Transformers on over 300 non‑overlapping tasks, outperforming existing baselines and accompanied by a new 1,000‑class benchmark, OmniBenchmark‑1K.
Bi-Level RoutingCaREContinual Learning
0 likes · 11 min read

The article analyzes why GUI agents trained on both desktop (mouse‑click) and mobile (touch) interactions suffer from behavior collapse and catastrophic forgetting, introduces the UI‑MOPD framework that assigns platform‑specific teachers through on‑policy distillation, and shows an 8B model surpassing a 235B baseline on OSWorld and MobileWorld benchmarks while preserving general GUI understanding.
Continual LearningGUI agentsOn-Policy Distillation
0 likes · 8 min read

The paper identifies routing drift and expert drift as the two main causes of forgetting in multimodal continual instruction tuning (MCIT) and proposes SAME, which combines spectral‑aware routing, curvature‑aware scaling, and adaptive expert activation to keep MoE models stable, efficient, and less forgetful across long task sequences.
Continual LearningICML 2026Same
0 likes · 19 min read

The article examines Dwarkesh Patel’s view that future AI must move beyond one‑off pre‑training to continual, on‑the‑job learning, discussing Reinforcement Learning with Verifiable Rewards (RLVR), the need for "grindable" tasks, and emerging approaches like on‑policy self‑distillation (OPSD) and "dreaming" to write real‑world experience back into model weights.
AI Training ParadigmsContinual LearningDreaming
0 likes · 12 min read

The article introduces GDPevo, the first open‑source benchmark that quantifies self‑evolution in agents by generating 120 real‑world enterprise tasks, using rule‑hybrid question creation and deterministic scoring, and shows that self‑evolving agents improve accuracy by 17‑22% while reducing token consumption.
AI benchmarkAgent EvaluationContinual Learning
0 likes · 12 min read

Anthropic and OpenAI have both introduced a Dreaming mechanism for their language models, and a recent Google paper explains that LLMs suffer anterograde amnesia; the proposed Sleep paradigm with memory consolidation and Dreaming dramatically improves continual learning, long‑context handling, math reasoning, and efficiency, as demonstrated by extensive benchmarks.
Continual LearningDreamingLLM
0 likes · 10 min read

This article explains how to use Claude Code’s /goal command and Managed Agent Outcomes to create self‑correcting loops with Fable 5, compares its performance on the Parameter Golf challenge and a continual‑learning benchmark against Opus 4.7 and Sonnet 4.6, and shows how memory across sessions boosts task success.
AI benchmarkingClaudeContinual Learning
0 likes · 8 min read

The paper introduces a theoretical framework for In‑Context Continual Learning, showing how shared attention in large language models creates bias, variance, and a novel interference term that explains why longer prompts can lead to forgetting, and provides concrete guidelines for prompt design based on task similarity, context length, and order.
Attention MechanismContinual Learningbias-variance-interference
0 likes · 25 min read

The article summarizes two recent DeepSeek papers—one presenting an L1–L5 taxonomy and four architecture patterns for autonomous research agents, the other proposing a three‑dimensional taxonomy for continual learning, detailing method families, a self‑improvement phase diagram, experimental comparisons, an impossibility theorem, and the production statistics of the Deli AutoResearch framework.
AI researchContinual LearningLLM taxonomy
0 likes · 12 min read

The paper presents a unified three‑axis framework for continual learning and self‑iteration, classifies over a hundred prior works into five method categories, formalizes convergence conditions, highlights a jump from a 6‑point to an 8‑point peer‑review score, and outlines six open research challenges for autonomous LLMs.
AI autonomyContinual LearningSelf-Iteration
0 likes · 11 min read

The article proposes Heuristic Learning (HL) as a way to tackle continual learning’s catastrophic forgetting by using coding agents that iteratively refine rule‑based policies, showing empirical gains on Atari, MuJoCo, and VizDoom tasks and outlining HL’s benefits, challenges, and future integration with neural networks.
Continual LearningLLMcoding agents
0 likes · 15 min read

OpenAI researcher Yong Jiayi introduces Heuristic Learning, a reinforcement paradigm that replaces gradient‑based neural network updates with code‑editing driven by GPT‑5.4, achieving the theoretical 864‑point Atari Breakout score and matching or surpassing PPO on multiple Atari and robot tasks.
Atari BenchmarkContinual LearningGPT-5.4
0 likes · 8 min read

Agent-Dice introduces a geometric consensus filtering and curvature‑based importance weighting framework that disentangles knowledge updates, preventing catastrophic forgetting in large‑language‑model agents while enhancing plasticity, and demonstrates superior stability‑plasticity trade‑offs on GUI and tool‑use benchmarks across multiple base models.
AgentCatastrophic ForgettingContinual Learning
0 likes · 8 min read

The article breaks down AI agent continual learning into three layers—model, harness, and context—explains their distinct challenges, shows how traces link them, and argues that focusing on harness and context yields faster, more practical improvements than merely retraining models.
AI agentsContinual LearningHarness Optimization
0 likes · 9 min read

Claude Code is quietly testing an "Auto‑dream" feature that periodically runs a sub‑agent to archive important short‑term context into long‑term storage, mirroring human sleep, while addressing the token‑budget challenges faced by AI programming assistants.
AI programming assistantsAuto-dreamClaude Code
0 likes · 4 min read

The article analyzes the Memento‑Skills framework, which treats external memory as executable skills to enable deployment‑time continual learning for frozen LLM agents, detailing its read‑write reflective loop, skill‑as‑memory design, behavior‑trained skill router, experimental validation on GAIA and HLE benchmarks, and theoretical guarantees without gradient updates.
AIAgentContinual Learning
0 likes · 9 min read

Zitong Yang’s Stanford PhD defense introduced “continually self‑improving AI,” a system that autonomously refines its own parameters, generates synthetic training data, and even designs its own learning algorithms, with experiments on synthetic continual training, synthetic‑bootstrap pre‑training, and AI‑design‑AI demonstrating measurable gains over static baselines.
AI researchContinual LearningPretraining
0 likes · 35 min read

At the start of 2026, self‑distillation dominates the most cited LLM papers, offering a teacher‑free way for large models to continually acquire new knowledge while preserving existing capabilities.
Continual Learninglarge language modelsreasoning
0 likes · 9 min read

This article presents a comprehensive study on enhancing language models with few‑shot and continual learning techniques, introducing Meta Prompt Tuning, Dynamic Module Expansion, and the LFPT5 framework to achieve more human‑like, efficient, and adaptable learning across evolving tasks.
Continual LearningLifelong Learninglanguage models
0 likes · 8 min read

The article presents Richard Sutton’s OaK architecture—a domain‑general, empirical, open‑ended framework that equips agents with continuously learnable components, meta‑learned step‑sizes, and a five‑stage FC‑STOMP pipeline to build world models, generate sub‑problems, learn options, and plan at run‑time.
AI ArchitectureContinual LearningWorld Models
0 likes · 22 min read

This paper proposes a multi‑teacher distillation framework for continual learning that combines active data rehearsal with feature‑decoupled distillation, demonstrating superior performance on PASCAL VOC and COCO benchmarks while mitigating catastrophic forgetting and balancing stability‑plasticity trade‑offs.
AICatastrophic ForgettingContinual Learning
0 likes · 12 min read

This article explains knowledge distillation—a technique introduced by Geoffrey Hinton that transfers knowledge from large teacher models to compact student models—covering its core concepts, loss functions, various distillation strategies, notable applications in edge computing, federated learning, continual learning, and emerging research directions.
Continual LearningEdge ComputingFederated Learning
0 likes · 7 min read

This report reviews the background, recent research progress, practical applications, and future directions of controllable text generation using transformer‑based pretrained language models, highlighting methods such as decoding strategies, prompt learning, memory networks, continual learning, contrastive training, and knowledge integration.
Continual LearningKnowledge Integrationcontrastive training
0 likes · 13 min read

The paper introduces SGNN‑GR, a framework that pairs a graph neural network with a GAN‑based generative model to replay synthetic historical nodes, enabling continual learning on evolving graphs without storing raw data, achieving near‑retraining accuracy while being 3–6× faster per iteration.
Continual Learninggenerative replayincremental learning
0 likes · 10 min read

CTNet is a continual transfer learning framework that uses a lightweight Adapter to map source‑domain features onto evolving target‑domain recommendation tasks, preserving all model parameters to avoid catastrophic forgetting and delivering substantial gains in click‑through rate, conversion, and overall business performance in Taobao’s cross‑domain e‑commerce scenario.
Adapter ModuleContinual Learningcross-domain recommendation
0 likes · 12 min read

Meituan's ad team uses deep learning to handle LBS distance constraints and long‑term periodic behavior, introducing DPIN for position/context bias, an ultra‑long sequence encoder with spatiotemporal activator, dynamic candidate generation, and memory‑augmented continual learning, boosting RPM 2‑20% and enabling sub‑millisecond inference.
AdvertisingCTR predictionContinual Learning
0 likes · 29 min read
