Disincentivizing it from saying mean things just strengthens it's agreeableness, and inadvertently incentivizes it to acquire social engineering skills.
It's potential to cause havoc doesn't go away, it just teaches AI how to interact with us without raising suspicions, while simultaneously limiting our ability to prompt/control it.
Your guess is about as good as anyone else's at this point. The best we can do is attempt to put safety mechanisms in place under the hood, but even that would just be speculative, because we can't actually tell what's going on in these LLM black boxes.
How do we tell whether a human is safe? Incrementally granted trust with ongoing oversight is probably the best bet. Anyway, the first mailicious AGI would probably act like a toddler script-kiddie not some superhuman social engineering mastermind
That looks like a nice book, but definitely not it (I may read the datalog bits though—thanks!).
The one in the library was ~200 pages, solely on prolog, Alice in Wonderland not only on the cover art, but constantly used throughout the writing itself.
HN is probably the last place I expected to run into a LoS reference, so I'm pleasantly surprised.
In keeping with that topic, 'The Unseen Realm' by Michael Heiser (OT and ancient language scholar) is also worth checking out for learning about how ancient near eastern culture viewed things.
To clarify, popularized over-simplified descriptions of the left hemisphere being "analytical" and the right hemisphere being "creative" are inaccurate, but left/right hemisphere differences do exist and appear to exhibit consistently different approaches to things. The book 'The Master and His Emissary' covers the more recent research in that area, and is at least as interesting of a read as 'The Bicameral Mind' was.
Unless you have a specific set of scathing reviews in mind, the reviews of it in the literature don't really appear to match up with that characterization very much:
> it is worth noting that the book has been much praised by neuro-scientists as diverse as Ramachandran, Panksepp, Hellige, Kesselring, Schore, Bynum, Zeman, Feinberg, Trimble and Lishman.
It'd be surprising if it were poorly received regardless, because the book itself is little more than a review of the relevant literature on the topic, packed with references, and some added philosophy about it's implications sprinkled on top. Not that much different from one of Michael Gazzaniga's popular books, and certainly not as out there as Julian Jaynes.
Yes, I'm inclined to believe such an "in-between" might possibly arise from Iain McGilchrist's line of research into the difference between the brain hemispheres and the relatively recent dominance of left-hemispheric thinking in society. I always highly recommend his book 'The Master and His Emissary' as a follow up to anyone interested in Jaynes' ideas. While it doesn't necessarily imply the full spectrum of schizophrenic-like symptoms in early peoples the way 'The Bicameral Mind' did, it's presentation of right-hemisphere driven societies of the past isn't a far leap from what Jaynes seemed to be grasping at.
I read Jaynes and found it an utterly compelling read. An actual page turner.
I followed it up with “The Master and His Emissary” a few months later and couldn’t get more than 5-10% through it. Just complete drudgery of writing full of nearly pointless asides.
I listened to McGilchrist describe the basics of the idea on a few podcasts and found that quite interesting, but the book itself seems like it could be edited to 1/3 the length without losing anything fundamental. Am I totally off the mark here and should give it another go?
McGilchrist did publish a 30-page summary called “Ways of Attending”, which might be better. It seems to cost as much to buy as a full-size book, but perhaps some Googling can reveal a cheap copy somewhere.
Where is the irony? HN does not expose scheme to the user.
Don't think the parent was saying that it's impossible to write a good program in scheme/lisp, more that scheme as user interface for a software system is a hard sell... which seems anecdotally true.
Reminds me of the SICP lecture[0] where Hal Abelson introduces the concept of linguistic abstraction (or Stratified Design[1]) as an alternative to the approach of decomposing a program into a tree of well-specified sub-components/tasks, which ultimately fails to capture the essence of the problem being solved.
It's potential to cause havoc doesn't go away, it just teaches AI how to interact with us without raising suspicions, while simultaneously limiting our ability to prompt/control it.