blutoot

Are "hallucination" from Google's "AI Summary" in Search Results very common?

Go to google.com and type in"how do I ignore AGENTS.md in codex?" --> the "AI Overview" section on the top of the search results page confidently says "To ignore AGENTS.md in Codex, you can use the codex --bypass-agents command-line flag to disable its loading, pass alternative instructions via flags, or leverage the optional AGENTS.override.md file for temporary control, but directly "ignoring" it often involves command-line overrides or creating higher-priority files rather than simple exclusion, as Codex is designed to find these instruction files by default." Then it goes on to give two examples:

In reality, it seems like Google's search AI (Gemini I suppose?) picked up the third search result on the front page which happens to be an open GitHub PR - https://github.com/openai/codex/issues/5983 - "Ability to bypass/disable AGENTS.md pipeline (CLI) #5983"

So, is the problem here that whatever is prompting Gemini is not sending the full context of the search results? (like in this case, the PR is actually open)

https://iceberg.mit.edu has got me spooked. If AI is going to be so disruptive to our way of life in such a short span of time causing immense economic pains to almost entire humanity, then every country's Government should take over and artificially control its progress. Civilian population should not be allowed to profit from it or spread it around even for free in the society. AI should be treated like a nuclear or a biological weapon. Countries should have non-proliferation treaties among themselves with perhaps mutually assured destruction.

Humanity's survival through a forced slow progress in automation is more important than being forced to "enjoy" the magic of automation. This isn't like replacing horses with cars.