LLM 35
- The Retry That Charged Twice: Idempotency Keys and Retry Budgets for LLM Calls
- The Cache Was Confident and Wrong: Choosing a Semantic Cache Threshold You Can Defend
- Same Script, Fifty-One Answers: What Morocco's Open Darija Release Actually Solves
- The Tokenizer Was the Point: Measuring MORENA, an Open 1.5B Model for 12 African Languages
- Guess Ahead, Pay Once: Speculative Decoding You Can Verify
- Half the Cache, Twice the Context — measuring what KV quantization costs at long range
- One Licence, Three Checkpoints: What the MiMo-V2.6 Open Release Hands Builders
- The Backdoor You Approved: MCP Servers as a Payments Attack Surface
- Follow the Leak: Four Channels a Prompt-Injected Agent Uses to Exfiltrate Data
- Global AI Roundup — August 2026
- DeepSeek V4 Peak/Off-Peak Pricing: The End of Flat-Rate LLM APIs
- How We Built a Skills-Powered Coding Agent with OpenCode (and What We'd Do Differently)
- Evaluating LLMs: Benchmarks That Matter for African Use Cases
- Fine-Tuning LLMs on African Language Datasets
- Building RAG Systems for Low-Resource African Languages
- How to Self-Host Open-Weight LLMs: A Practical Guide
- Kimi K3: China's 2.8 Trillion Parameter Challenge to Silicon Valley
- Agent Observability and Debugging: Tracing, Logging, and Monitoring
- Building Secure AI Agents: Guardrails, Validation, and Sandboxing
- Deploying AI Agents to Production: Scaling, Error Handling, and State Management
- Multi-Agent Systems: Orchestrating AI Teams with LangGraph and CrewAI
- Tool Use and Function Calling: Giving Agents Real-World Capabilities
- Memory Systems for AI Agents: Short-Term, Long-Term, and Persistent Storage
- Building AI Agents from Scratch: Fundamentals, Tools, and the ReAct Pattern
- Real-Time Multimodal Security: Attacks via Audio and Video
- Automated AI Red Teaming at Scale
- Prompt Injection via Structured Data Formats
- Memory as an Attack Surface: Poisoning AI Agent Memory
- Supply Chain Attacks on AI Systems: From Model Repos to Pipelines
- Insecure Agent Design: When AI Has Too Much Agency
- Data Poisoning and Model Backdoors: Training-Time Attacks on AI
- Jailbreaking LLMs: From DAN to GODMODE
- Prompt Injection: The #1 LLM Security Risk
- Knowledge Graphs Meet LLMs: RAG with Structured Knowledge
- RAG Security: The Hidden Attack Surface of Retrieval-Augmented Generation