13 ms·

Sketch-of-Thought: Efficient LLM Reasoning

• azaras 2y ago
  It is interesting for an old and classic software developer to see how the LLMs are taught/programmed. They use markdown files: - https://github.com/SimonAytes/SoT/blob/main/sketch_of_thought/config/prompts/EN/EN_CoT_SystemPrompt.md https://github.com/SimonAytes/SoT/blob/main/sketch_of_though... - https://github.com/SimonAytes/SoT/blob/main/sketch_of_thought/config/prompts/EN/EN_ConceptualChaining_SystemPrompt.md https://github.com/SimonAytes/SoT/blob/main/sketch_of_though... - https://github.com/SimonAytes/SoT/blob/main/sketch_of_thought/config/prompts/EN/EN_ExpertLexicons_SystemPrompt.md https://github.com/SimonAytes/SoT/blob/main/sketch_of_though...
  • jxjnskkzxxhx 2y ago
    If it fullfills the purpose, then it's a good enough solution.
• Bolwin 2y ago
  This seems near identical to chain of draft, albeit more thorough and they actually did training instead of just promoting https://arxiv.org/abs/2502.18600 https://arxiv.org/abs/2502.18600
  • xyz7786 2y ago
    After reading both papers, I have to disagree. Chain of Draft seems "identical" to Concise Chain-of-Thought (CCoT) https://arxiv.org/abs/2401.05618 https://arxiv.org/abs/2401.05618. Additionally, the CoD authors seem to mention that it is cognitive inspired but don't give any actual basis for that. Sketch-of-Thought, while similar in terms of its goal, seems to be based on cognitive science principles and has better extensibility than CoD and CCoT (they basically just say "answer in X words" and leave it at that).