WheelyCurious wrote: Sun Dec 18, 2022 5:39 am
I subscribe to a free computer security letter put out by Bruce Schneier who is one of the top level experts on cryptography and computer security, mostly summarizing and giving links to the various security bugs, exploits etc. in the online world...
In his latest issue, he posted the result of a test he did on one of the AI engines where he told it to "write an essay on AI in the style of Bruce Schneier" - It did a pretty good job, I might have bought it as his.... His own comment was that it wasn't at the point he could retire, but he might be able to go on vacation for a few days... Scary for sure....
WheelyCurious
And totally not scary. AI at this stage doesn't know when to stop, and if it lacks details, it confabulates 'believable bullshit'. (coworkers tried asking it questions for IT job interview. Results were "believable on the surface, confident and wrong"). Worse - there is no indication - when it was writing something true and checked and when it was total fiction.
So - if you need precision and not fiction - all what was generated needs to be human-reviewed. And as it was not human who generated it - it will/may require more time.
IMO, this will fill up internet with more advertisements and stuff, indistinguishable from some sane text at first glance and it will make finding something specific even harder than it is now.
At least before Artificial Idiot - you'd need to pay humans. Now you can do it essentially for free.
To quote another one:
- Hey, John, come back to us, we need a programmer!
- But you have monkey with GPT-3 AI.
- Monkey with GPT-3 AI is now a team lead manager. We need programmer again.