Ah, it's been so long since my last post! Today, I'll review some HOT๐ฅ ๐๐ฟ๐ฎ๐ฝ๐ต-related concepts. Broadly, we'll explore them in this order:
- ๐๐ป๐ผ๐๐น๐ฒ๐ฑ๐ด๐ฒ ๐๐ฟ๐ฎ๐ฝ๐ต ๐
- ๐๐ด๐ฒ๐ป๐ ๐๐ฟ๐ฎ๐ฝ๐ต ๐บ๏ธ
- ๐๐ฟ๐ฎ๐ฝ๐ต ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด โ๏ธ
1. ๐๐ป๐ผ๐๐น๐ฒ๐ฑ๐ด๐ฒ ๐๐ฟ๐ฎ๐ฝ๐ต
- This is the 'knowledge graph' that has been around for a long time. It used to be built with tools like neo4j. The reason this knowledge graph has become hot again is likely due to the undeniable synergy of Karpathy's ๐๐๐ -๐ช๐ถ๐ธ๐ถ's influence + the keyword ๐ข๐ป๐๐ผ๐น๐ผ๐ด๐.
- In my personal opinion, llm-wiki has contributed to lowering the burden associated with the concepts of 'knowledge graph' and 'Ontology,' which previously felt distant, by enabling quick utilization and visualization, making them 'easy for people to understand.'
- If you haven't tried it yet, I recommend the video below. It's short, clear, and concise. https://www.youtube.com/watch?v=iXd0t60YmMwyoutube.com
2. ๐๐ด๐ฒ๐ป๐ ๐๐ฟ๐ฎ๐ฝ๐ต
- These concepts focus on 'structuring a Repo as a Graph' so that an Agent understands the current Repo well and both AI/humans can understand it easily. They are also a type of knowledge graph.
- Popular examples these days include ๐จ๐ป๐ฑ๐ฒ๐ฟ๐๐๐ฎ๐ป๐ฑ-๐๐ป๐๐๐ต๐ถ๐ป๐ด, ๐ฐ๐ผ๐ฑ๐ฒ-๐ฟ๐ฒ๐๐ถ๐ฒ๐-๐ด๐ฟ๐ฎ๐ฝ๐ต, and ๐ด๐ฟ๐ฎ๐ฝ๐ต๐ถ๐ณ๐. You can choose one based on its specific features.
- By doing this, it's like handing the AI a map, which prevents it from wasting tokens by rummaging through irrelevant code even for a simple modification. It will directly find the correct place to make changes. And clear explanations for me are a bonus!
3. ๐๐ฟ๐ฎ๐ฝ๐ต ๐๐ป๐ด๐ถ๐ป๐ฒ๐ฒ๐ฟ๐ถ๐ป๐ด
- After PROMPT, CONTEXT, and LOOP, now it's Graph Engineering. haha (Who decides these things anyway..๐ซ )
- Actually, it's not such a new concept, but I believe it's a term that became established after someone's idea of "What if we place LLMs as nodes in a Graph, and then create a supervisory LLM to oversee them all?!๐" gained a lot of agreement. Even though it's something that has often been done with tools like LangGraph, n8n, etc. Just as society develops and we get hoes, shovels, pickaxes, cultivators, tractors, and even drones for spraying pesticides, all you need is the right tool, even for farming.
- By utilizing each concept well, it would be good to simply avoid situations where 'what could be stopped with a hoe ends up needing a shovel' (meaning, preventing small issues from escalating).



0 comments
Please login to leave a comment.