Downloading the archive of my tweets just so I can actually search them... turns out it also includes the full text of every tweet I've ever liked?!
(That hasn't been deleted since, I guess I'll have to start taking snapshots and adding them up.)
The real blackpill is that it doesn't matter if you lie. Might even be a good plan, if you're length constrained (orange markers) (Figure S4)
Or maybe that's a whitepill. If you're telling 80% lies, surely you're well past the point where Gell-Mann is a good defense.
New w/ @AISecurityInst & @UniofOxford:
Frontier AI can now out-persuade expert humans in conversation - incl. world-champ debaters and professional canvassers.
This held even when humans chose their topics, prepared in advance, and competed for £1,000 prizes 🧵
So a weird thing is that I haven't seen anyone suggest that a journal that cared would do well to bar not just those caught, but those who publicly argue to excuse fake citations.
1) Feb: I read a paper about 51% attacks on PoW chains and decide it has a rather optimistic assumption.
xitter.cf/BoilingPb/status/20195…
Replying to @planefag @NotWoofers
This paper: 'if you let anyone rent your army, it would be very cheap for someone to pay 51% of it to shoot it out with your 49% for a couple minutes...'.
Which is true, but like with militaries, only a couple percent of hashpower is even on the market. (Though see 5.4.1/2)
2) Mar + April: an amusing parallel between a heavy metal and the Christian theology of sin.
xitter.cf/BoilingPb/status/20322…
Replying to @blind_via @open_erv
I shouldn't speak for others, but the pun is too good...
I think he just doesn't let original sin (every igneous rock contains at least 10 µg/g), or hydraulic transport (by which concentrations in many environments are decreased), excuse relativism or nihilism or nonchalance.
#0 For ~the same reasons as before, I'm rebooting the scheduled low-quality top-level-posting exercise.
Until further notice, every Saturday, there will be a new post below... if it's stupid, opaque, or cringe, kindly assume that's intentional.
#35 any relation to contemporary discourse is accidental.
xitter.cf/nickcammarata/status/1…
#36 They say what gets measured gets managed but it doesn't really look like it's working.
In the replies people blame demographics, but what if that's downstream of structure?
What if 4chan (etc.) were so powerful because they combined light moderation, bottom-posting, non-branching threads, thead-bumping, and trivial (but nonzero) border policy?
Apparently the full Wikipedia text dump is ~10TB compressed, because they just keep the full article text, including xml headers, for every edit.
(For guessing at uncompressed size—the current text of all articles is ~25GB compressed / 110 GB uncompressed.)
If LLMs' preference for certain constructions exist because they help them think, does avoiding them make *you* dumber—at least when you're constructing text linearly?
Replying to @kitten_beloved
Perhaps it makes them smarter?
'not X. Y.' seems like simply the most compact way to basin hop / narrow.
And the em-dash is the high-quality-text–coded representation of an optional-return, optionally grammatical, abrupt aside.
Dylan Foltz retweeted
If there are any cracked, young engineers, maybe college students, who might want to build this part time, I will pay you to build this.
The "R9X drone" / interceptor (and EDF version) must be built. 🚀 I won't wait for angels/VCs to get this going. RT for vis appreciated. 🙏
> 'I'll get a blue light filter for my monitor, then I can use the machine more late into the night.'
> it works
> (gets sleepy late at night)
To show red-on-white (like spellcheck underlines) I'm running in the default inverted greyscale, though given how visible straight green is a bichrome might be possible?
Unhappy with the rendering, BTW. Hard to convey the sort of orange a ~630nm "red" is...



















