Re the LLM issue, asking the free ones on DuckDuckGo Chat the question "What init system does GhostBSD use?" gets several wrong answers:
- GPT-4o mini: "GhostBSD uses the OpenRC init system. ..."
- Claude 3 Haiku: "GhostBSD, a FreeBSD-based operating system, uses the OpenRC init system. ..."
- Mistral Small 3: "GhostBSD uses the OpenRC init system. ..."
Note GPT-4o mini is trained on data up to October 2023, and Claude Haiku to August 2023, so about two years after GhostBSD switched back from OpenRC. Better answers came from:
- Llama 3.3 70B: "GhostBSD uses the RC init system, which is also used by FreeBSD."
- o3-mini: "... GhostBSD uses the FreeBSD rc(8) init system and its accompanying rc.d scripts for service management."
Llama 3.3 70B was trained on data up to December 2023, and o3-mini to October 2023, so the issue here isn't the recency of the data, but the quality.
Given how many people are scarily reliant on LLM advice, it seems a good idea to get outdated training data removed from circulation.