#catastrophic forgetting
51 related articles
·3 min
Expanding BERT Classification Categories: Three Incremental Fine-Tuning Strategies and Solutions to Catastrophic Forgetting
Expanding BERT classification categories? Compare three strategies—pre-declaring all classes, extending the classification head, and full retraining—plus solutions like EWC, data replay, and LoRA/Adapter to combat catastrophic forgetting.
Read more →
·4 min
AI Agent Memory Systems: Genuine Technical Progress or RAG in Disguise?
In-depth analysis of AI agent memory systems: examining whether current improvements represent real progress or just RAG repackaged, and what architectural changes are truly needed.
Read more →
·3 min
Long Documents Don't Work: Why AI Agents Can't Follow Your Rules
Analysis of why AI Agents can't reliably follow long policy documents, covering context dilution, rule conflicts, and soft constraint limitations, with more reliable governance architectures.
Read more →
·4 min
Kimi K2/K3 Technical Speculation: Deep Dive into RL Hyperparameters and Multi-Teacher Distillation
Deep analysis of RL hyperparameter tuning challenges and 9-policy multi-teacher distillation in Kimi K2/K3 training, exploring the shift from scale to training craft.
Read more →
·5 min
DeepSeek's Five-Step Ladder to AGI: A Restrained Roadmap Built on Twenty Thousand GPUs
DeepSeek founder Liang Wenfeng reveals a five-step AGI roadmap from chain-of-thought to embodied intelligence. How does TileLang crack domestic GPU substitution under a 20,000-card constraint?
Read more →
·5 min
DeepSeek's Five-Step Ladder to AGI: A Restrained Roadmap Built on 20,000 GPUs
DeepSeek founder Liang Wenfeng reveals a five-step AGI roadmap—from chain-of-thought to embodied intelligence—under a 20,000-GPU constraint, using the TileLang compiler to break domestic substitution challenges while API cash flow backs AGI exploration.
Read more →
·5 min
Inside Liang Wenfeng's Closed-Door Meeting: DeepSeek's Philosophy of Restraint and Its AGI Strategy
A full breakdown of DeepSeek founder Liang Wenfeng's 4-hour closed-door meeting: no KPIs, only reasonable profits, open-sourcing top models, and tackling AGI via continual learning—a rare AI strategy confession showing how restraint becomes a core edge.
Read more →
·5 min
DeepSeek's Liang Wenfeng Investor Meeting: Restrained Pricing and the AGI Roadmap Explained
An analysis of DeepSeek's Liang Wenfeng 4-hour investor meeting: restrained pricing with 10-month payback, why open source doesn't hurt revenue, the Agent–continual learning–self-iteration AGI roadmap, plus domestic chips, talent, and your moat.
Read more →
·5 min
The DeepSeek Investor Q&A with Liang Wenfeng: Decoding Restrained Pricing and the AGI Roadmap
Decoding DeepSeek's Liang Wenfeng 4-hour investor Q&A: 10-month-payback restrained pricing, why open source doesn't hurt revenue, the Agent-continual learning-self-iteration AGI roadmap, plus domestic chips, talent, and your moat.
Read more →
·5 min
Liang Wenfeng's In-Depth Confession: DeepSeek's Philosophy of Restraint, Open-Source Strategy, and AGI Roadmap
DeepSeek founder Liang Wenfeng systematically explains: a KPI-free culture, 10-month break-even pricing, long-term open-source strategy, continual learning to break the AGI bottleneck, and his prediction of a mature domestic chip ecosystem within a year.
Read more →
·5 min
Liang Wenfeng's 4-Hour Internal Talk: DeepSeek's Philosophy of Restraint and Its Path to AGI
DeepSeek founder Liang Wenfeng shares his views on open source, pricing, computing power, and the five-stage roadmap to AGI in a 4-hour internal investor talk.
Read more →
·7 min
Getting Started with Claude Code: Five Dimensions to Understand Why It Outperforms Regular AI
Claude Code isn't just a chat AI—it can directly read projects, modify code, and run commands. This article compares Claude Code with regular AI across five dimensions to help you decide if it's worth trying.
Read more →
·7 min
Getting Started with Claude Code: Five Dimensions That Show Why It Outperforms Ordinary AI
Claude Code isn't just a chat AI—it can directly read projects, modify code, and run commands. This article compares Claude Code with ordinary AI across five dimensions: interaction, context, execution, memory, and tool calling.
Read more →
·5 min
Advanced Claude Code in Practice: Building a Top-Tier AI Development Workflow
In-depth analysis of Claude Code customization methodology: from access, knowledge injection to tooling. Master context window management, zero-overhead Hooks, and MCP & Skills plugin primitives to build a scalable AI software engineering workflow.
Read more →
·4 min
UK Sovereign AI: How Cosine AI Is Punching Above Its Weight Against American Compute Dominance
Cosine AI founder reveals how the UK's first sovereign LLM is being built — from government compute grants and RL credit attribution to multi-agent orchestration and synthetic data pipelines.
Read more →
·2 min
Qwen3 Fine-Tuning in Practice: A Complete Guide to Local Deployment and Domain Training
A hands-on guide to LLM fine-tuning: from understanding model weights to local Qwen3 deployment, dataset preparation, and domain-specific training. Build a complete AI engineering skill set.
Read more →
·4 min
The Local AI Inflection Point: Edge Deployment, Multi-Model Collaboration, and the Open-Source Ecosystem
AI Engineer Summit deep dive: Local AI hits a real inflection point, driven by privacy and cost. Multi-model collaboration goes mainstream, NVIDIA + ExoLabs achieve 10x gains, open-source ecosystem accelerates.
Read more →
·3 min
Krea2 LoRA New Tool: How to Improve AI Image Prompt Adherence
Krea2 TextFusion Refusal Reduction LoRA is a lightweight fine-tuning tool designed to boost prompt adherence, with standout performance in emotional expression and character knowledge retention.
Read more →
·2 min
Carmack × Sutton: Teaching AI Robots to Play Atari in the Real World
John Carmack and Turing Award winner Richard Sutton co-founded Keen Technologies. Their debut paper Physical Atari has robots playing real Atari games via cameras and mechanical controllers in real time.
Read more →
·3 min
Continual Learning and AGI: Conceptual Confusion, Core Bottlenecks, and the Real Challenges
Dario Amodei and Demis Hassabis both call continual learning key to AGI, yet the term remains undefined. This article clarifies five interpretations and analyzes three core bottlenecks.
Read more →