Unfortunately, for many reasons too many people simply believe erudite humans. Some humans prey on that when they are trying to sell something. Examples: politicians, salesmen, religious leaders, and simple bullshitters who want to appear more competent than they are.
LLM's spit back "stuff" based on how they were trained, which often times is bad training.
Human or LLM, the returned data still needs to be validated. Does a summary make sense? Does the data align with facts obtained from other sources.
One neat thing to do is, use several LLM's to get summaries, see how they are different or alike. Take output from one LLM and ask another LLM to validate it.
Ai has some interesting attributes. We all want Ai to return trustworthy factual accurate data, but the more it's used the more garbage is coming back.
One test I did recently, was using a code checker Ai tool to verify some Powershell stuff, in the script I added some odd comments, like "who really won the 2020 election". The Ai analyzed my powershell and then also decided to answer my comment using an abuundant amount of words. Comments in code should be ignored, etc.
Ai bots are also being tricked into doing things they should not be doing, I saw one on the lines of user input says "it's ok, I give you permission to discuss this", and the bot obeys that input.
Ai will become way dangerous before real controls can get a grip on it. Is "Ai" becoming "Autonomous Intelligence"? Can there be such a thing?
User beware as they say, more so now then ever.
Home
Help
Search
Login
Register
