- Interviewed on 2024-03-04
- Currently works at a GPT Project
- Sunghyun Cho is a senior software engineer with a background in AI, frequently uses GPT for both automation and personal purposes.
- Experiences diminishing quality in GPT responses, prefers older GPT-4 versions via MindMac for better results.
- Developing Heimdall, a Hackernews summarizer tool using GPT.
- Plans to join a medical AI company.
- Faces challenges identifying problematic prompts when using GPT.
- Seeks a tool for iteratively developing and testing prompts in one platform.
- Envisions Helicone as a solution for independent developers to address the complex entry into "LLM-Ops".
- Desires a more accessible, cleaner, and modern alternative to Langchain for LLM operations, emphasizing easy streaming capabilities.
- Prefers a library for LLM Ops not owned by Vercel, featuring streaming, JSON modes, and regex mode.
- Criticizes RAG (Retrieval-Augmented Generation) for its simplicity and limited effectiveness.
- Current GPT iterations lack depth in conversations, with perplexity being the only notable metric for learning new concepts.
- Notes language models' limitations with non-Roman characters, such as Korean.