@willdepue@george_veng@space_colonist The thing to watch for is how effectively we attack this constant factor. Natural language is an extremely inefficient means of auto-regression.
If LLMs start using efficient representations of internal state for auto-regression, my timeline estimates change dramatically.
@willdepue@george_veng@space_colonist ... but, does that mean that gradients can't carry us to brain-equivalency and beyond? No, that's too strong a conclusion.
Auto-regression makes DNNs and RNNs computationally equivalent. It's a giant constant factor, asymptotic complexity is the same.
@WesternMishima@pegobry_en The fact that they got that right and still got the big conclusions wrong about what it would mean is actually more damning for their credibility, not less.
If they wanna drop an article about SoTA capability evolution, cool, I'm all ears but this is eschatology fanfic
and like here's the thing
I can say "a lot of their scenarios sound implausible and outlandish to me" and that's true
but I don't actually have a competing plausible and landish scenario in mind, I just have a giant question mark in a world undergoing massive changes
@FlawlessContext@CharlieBull0ck But it *is* fiction, by definition, layered over some concrete, trackable predictions (AI 2027 made a major point of this, with mixed accuracy so far).
The narrative portions themselves are SciFi and the authors themselves note there's a ~0% probability it plays out exactly.
@AgainstWeaves@VoidScreamr@downbadcomment You're implying that it's fine for love to fade if there's kids involved.
But in fact, if love fades and there's no kids, separation is a clean solution with no collateral damage.
Despite best efforts, not all unions will last.
George III: I just wanted you to know that you can't just say the word "independence" and expect anything to happen.
Jefferson: I didn't say it. I declared it
@CapitalShipyard@wonderingbread1@saylor I asked my Claude to pull the options chain and build a model. The upshot is that they're about the same; the indifference point is about 1x NAV. I suspect StatArb will prevent Strategy from falling much below 1x.
I'm not a buyer here but if it was 0.8x I would be.
@CapitalShipyard@wonderingbread1@saylor The discount to NAV is generally because these capital allocators take huge management fees (2 and 20, possibly). That doesn't really apply here