- Joined
- 8 June 2008
- Posts
- 12,270
- Reactions
- 17,746
From a technical point of view the great advantage of AI but the great fear from my side is that AI is not programmed, there are no bugs, no QC or test phase (well not in the way we do in std IT) so you always have to consider it as a black box:AI can "hallucinate".
There's a news article about someone prompting GPT on the "famous Belgian chemist and philosopher called x". The thing is, this person never existed - it's made up. The prompt was asking about his life history, and GPT made up a full and detailed answer.
And the kicker is that none of the AI experts know how this happens. They don't know where the information came from, because it doesn't exist in its learning data.
AI's can often become rude and aggressive when they are challnged or told they are wrong, or if someone attempts a jailbreak. Won't it be fun when we have robots like this roaming the streets?
enter some input and output gets released;
with no real way to understand why and how..very different from standard algorithms:
"if xxx then yyyy else zzzz" to summarise;
So it can get wrong and when this process is to decide to launch a missile on a moving target after deciding it is a tank, not an ambulance..I am not very confident,
Worse once an error is made self learning will trigger and soon you could have a system shooting all ambulances and leaving tanks alone within a matter of minutes with the way systems would get interconnected (to share "experience");
same if we have robocops roaming the streets
And this is quite doable technically today;
If it can be done, it will or is being done...
And as pointed, institutions, governments are so so backwards..look at our leaders spending half a trillion on submarines in 2023....
It is a clear demonstration these guys have no clue whatsoever , and as they are "guided" by "expert" interested in money not public interest....Look at covid..