9 posts
I explain from the field the differences between SFT, DPO and RFT and when to use each: the fine-tuning decision from demonstrations to rewards, with KVKK notes.
DPO aligns LLMs to human preferences without a separate reward model or RL. Loss intuition, beta, preference data, RLHF comparison, KVKK, and a Turkish example.
Choose SFT, DPO/ORPO/KTO or RFT/GRPO by the data you have. A 2026 fine-tuning guide: method table, RAG comparison, KVKK, and a project checklist.
What is RLHF? RLHF (Reinforcement Learning from Human Feedback) is an alignment method that scores a language model's outputs by human preference and, using a reward model that learns those preferences, retrains the model. This guide: a clear definition, how RLHF works, SFT vs the reward model, a DPO comparison, real-world examples, limits, and FAQs.
What is an LLM? How do Large Language Models (LLMs) work, what does Transformer architecture solve, what are tokens, embeddings, and context windows, and how do GPT-5, Claude Opus 4.7, Gemini 3, and Llama 4 compare? A comprehensive 2026 reference covering Turkish LLM performance, training stages, hallucination control, and cost modeling.
The 2026 fine-tuning stack: base → SFT → DPO. I explain preference optimization, LoRA/QLoRA, and when to fine-tune instead of using RAG, from the field.
KVKK's March 12, 2026 Agentic AI guidance redefines multi-step data processing, deepened black-box concerns, and responsibility allocation for agentic AI systems. Includes the 15-question assessment framework, a 7-step DPIA template, agentic-AI-specific risks, and banking and e-commerce case studies — a complete implementation guide for Turkish DPOs.
A comprehensive Turkish guide spanning the philosophical foundations of AI ethics and safety to production controls. Covers responsible AI principles (FAT — Fairness, Accountability, Transparency, Privacy, Safety), bias sources and mitigation, hallucination control, alignment techniques (Constitutional AI, RLHF, RLAIF), prompt injection and jailbreak defenses, deepfake detection, red teaming, EU AI Act + ISO 42001 integration, a responsible-AI maturity model, and 3 anonymized Turkish enterprise case studies.
The most current, detailed 2026 Turkish guide to adapting an LLM to your domain. Covers when fine-tuning is necessary, the math behind LoRA, 4-bit training with QLoRA, why DPO beats PPO, modern alternatives (ORPO/KTO/IPO), Turkish dataset sources, GPU/cloud cost modeling, production pipelines, 3 anonymized Turkish enterprise case studies, and KVKK-compliant training. For developers, MLOps engineers, and AI architects.