Hugging Face Journal Club - DeepSeek R1
- Source
- youtube.com
- Author
- Hugging Face
- Date
Context
DeepSeek's R1 work separates learning to solve checkable problems from learning to produce usable answers. The journal-club discussion explains how R1-Zero rewarded correct outcomes on tasks such as mathematics and coding, starting from an already pretrained model without an initial supervised stage. Watch the discussion.
The researchers' own model summary reports that R1-Zero developed reasoning behaviors but also repeated itself, mixed languages and produced hard-to-read output. R1 added example-based training and further reinforcement learning to address those problems. The distinction matters because better problem-solving scores alone do not establish that a model communicates well enough for everyday use. Read the researchers' summary.
- fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
- RLHF — Reinforcement learning from human feedback — training a model to prefer answers humans rate as better, which turns a raw text predictor into a usable assistant.
videoDeepSeek, Reasoning Models, and the Future of LLMsa16z
article[AINews] DeepSeek v4.1-Flash: 763B-P8B-D16B novel causal Encoder–Decoder architecture with vision marks the Return of the Whalewww.latent.space
videoThe Unreasonable Effectiveness of Reasoning Distillation: using DeepSeek R1 to beat OpenAI o1Latent Space
Checking sign-in…
Loading comments…



