Coding & AI Resources

5 Fun Papers That Explain LLMs Clearly 1️⃣ Attention Is All You Need 📝 Description: Introduced the Transformer, the architecture behind every modern LLM. Replaced older recurrent/convolutional models for sequences. 🔑 Key Ideas: Self-attention • Multi-head attention • Positional encoding • Transformer block 🔗 Paper: https://arxiv.org/abs/1706.03762 ━━━━━━━━━━━━━━━ 2️⃣ Language Models Are Few-Shot Learners 📝 Description: The GPT-3 paper. One 175B model handles many tasks just by reading prompts — no retraining. 🔑 Key Ideas: In-context learning • Few-shot prompting • Autoregressive next-token prediction 🔗 Paper: https://arxiv.org/abs/2005.14165 ━━━━━━━━━━━━━━━ 3️⃣ Scaling Laws for Neural Language Models 📝 Description: Showed model performance improves predictably as parameters, data & compute grow. The logic behind going big. 🔑 Key Ideas: Scaling laws • Compute-optimal training • Data vs. model size tradeoffs 🔗 Paper: https://arxiv.org/abs/2001.08361 ━━━━━━━━━━━━━━━ 4️⃣ Training LMs to Follow Instructions with Human Feedback 📝 Description: The InstructGPT paper. Turns a raw text predictor into a helpful, instruction-following assistant. 🔑 Key Ideas: RLHF • Supervised fine-tuning • Reward model • Human preference ranking 🔗 Paper: https://arxiv.org/abs/2203.02155 ━━━━━━━━━━━━━━━ 5️⃣ Retrieval-Augmented Generation (RAG) 📝 Description: LLMs fetch external documents instead of relying only on stored memory — great for facts that change over time. 🔑 Key Ideas: Dense retrieval • Document index • Grounded generation • Knowledge-intensive QA 🔗 Paper: https://arxiv.org/abs/2005.11401 ❤️ Follow  for more