AI Papers — Key Research Made Accessible
The most important papers on LLMs, RAG, Agents and AI Safety — summarized and explained.
Attention Is All You Need (2017)
The Transformer paper: Why Self-Attention changed the entire AI landscape.
Retrieval-Augmented Generation (2020)
RAG explained: How LLMs become better and more reliable through external knowledge sources.
LoRA: Low-Rank Adaptation (2021)
Parameter-efficient fine-tuning: Adapting large models without retraining everything.
ReAct: Reasoning and Acting (2022)
The ReAct agent pattern: How LLMs solve tasks by alternating between thinking and acting.
Constitutional AI (2022)
AI Safety by Anthropic: How to align AI systems through principles instead of human feedback alone.
11 more articles (generated)
Generated in March 2026, not editorially reviewed — every article says so in its header.
BERT — Bidirectional Encoder Transformers
Masked Language Modeling, Next Sentence Prediction, bidirectional context, fine-tuning paradigm, impact and legacy (RoBERTa, ALBERT, DeBERTa).
2026-03-21 · generated, not reviewed
Chain-of-Thought — Step-by-Step Reasoning
Wei et al. 2022, Few-shot CoT prompting, Zero-shot CoT ("Let's think step by step"), Self-consistency, Tree of Thoughts extensions.
2026-03-21 · generated, not reviewed
DPO Explained — Direct Preference Optimization
"Direct Preference Optimization" (Rafailov et al. 2023). DPO is faster and cheaper than RLHF. Does it replace RLHF entirely?
2026-03-21 · generated, not reviewed
Flash Attention — Efficient Transformer Inference
IO-aware attention, tiling, memory hierarchy. Flash Attention 1 vs. 2 vs. 3, speed/memory improvements, practical inference impacts.
2026-03-21 · generated, not reviewed
LLaMA — Meta's Open-Source Language Models
LLaMA 1 through LLaMA 3. Training strategies, architecture decisions, RLHF in LLaMA 2, and why Meta open-sourced these models.
2026-03-21 · generated, not reviewed
Mixture of Experts Explained
"Mixtral of Experts" (Mistral 2024) + MoE concept. Sparse activation, routing, efficient large models. DeepSeek MoE and beyond.
2026-03-21 · generated, not reviewed
RLHF Explained — Training with Human Feedback
"Training Language Models to Follow Instructions with Human Feedback" (InstructGPT). How RLHF makes models useful and safe. Reward models and PPO training.
2026-03-21 · generated, not reviewed
Scaling Laws Explained — Compute-Optimal Training
"Scaling Laws for Neural Language Models" (Kaplan et al. 2020) + Chinchilla. How many parameters? How much data? The math behind large models.
2026-03-21 · generated, not reviewed
Stable Diffusion — Latent Diffusion Models
Latent Diffusion Models paper, VAE + U-Net + CLIP architecture, conditioning. SD 1.5 → SDXL → SD3 evolution.
2026-03-21 · generated, not reviewed
The GPT Series — From GPT-1 to GPT-4
The evolution of GPT-1 through GPT-2, GPT-3, and GPT-4. Architecture evolution, scaling laws, and why this series revolutionized the AI market.
2026-03-21 · generated, not reviewed
Whisper — Multilingual Speech Recognition
Radford et al., weak supervision at scale, encoder-decoder transformer architecture, multilingual training, model sizes, accuracy vs speed tradeoffs.
2026-03-21 · generated, not reviewed
Continue the learning path
The learning path puts these articles in order, and the Hub carries the building blocks we have checked in our own operations.
- Local and self-hosted
- Documented and verifiable
- From our own operations
- Made in Austria