Papers

• Foundation models require rethinking reliability

Articles + demos

• Building llama with python from scratch
• An amazing text classification demo with Distillbert for classifying food from non-food (code)
• https://huggingface.co/spaces/huggingface/inference-playground?modelId=deepseek-ai/DeepSeek-R1-Distill-Llama-70B
• https://tiktokenizer.vercel.app/

Summary of the Key points of the LLM deep dive by Andrew Karpathy

LLM as a judge

Tool use by large model

• you need data to show the model how to do search
• some examples to show the model how it does search and when to do search

Handling hallucination

• ask the model to use tools like code
• ask the model to conduct intermediate steps. like break the problem into small pieces.