zero

naraka enjoyer

[zero @appelbolt]

I had a few bangers for sure

[zero @appelbolt]

This materialized quickly

[zero @appelbolt]

@robertwiblin What he’s actually doing is threatening an organization he sees as the enemy.

[zero @appelbolt]

Hypervigilance is generally maladaptive. Locking key systemic variables far out of homeostatic optima is bad everywhere else and the same is true for meditative parameters like focus, introspection and vigilance. ‘Nothing in excess’ is literally an oracular truth.

[Θωμᾶς del Vasto @Thomasdelvasto_]

🔒 🔁 The phenomenon of serious meditatooors obtaining a hypersensitive CNS and immune system, as well as gut issues, make it sound like they're speedrunning autism

[zero @appelbolt]

@00Dazzle @nabeelqu Looks like neither from the game - it plays like a weak-intermediate human. Solid Ruy Lopez / positional play into a bad blunder is a very typical pattern at low intermediate human elo.

[zero @appelbolt]

@BjarturTomas In some ways worse imo (ie even more slop shape) which is an odd thing to coincide with all the benchmark breaking

[zero @appelbolt]

@meaning_enjoyer I’ve had the same thought many times and as a result try to give it sincerely as advice whenever I can

[zero @appelbolt]

@reconfigurthing I find it tough to read now, even Fable. Shot through with Claude-isms and Claudish cliches. 4.7 would load documents with extraneous cruft. I found 5.5 a much better editor and its proposals generally more elegant/restrained therefore more usable.

[zero @appelbolt]

It doesnt really seem like a good thing. I also noticed it from 4.6 onwards as an unpleasant combination of verbosity and compression. Its tough to argue with every benchmark ending up smashed but given the territory a progressive decrease in legibility seems like it should at least raise concerns.

[zero @appelbolt]

@zetalyrae Almost always, I agree. But I think irregular progress happens in it sometimes anyways, behind or beneath the object level conflict. Debates are also (eg ancient Athens) maybe more often about the audience than the opponent.

[not lydia @notlydiabro]

@SchrodingrsBrat

[zero @appelbolt]

@asher5772 you can do this with tmux if you just want cross-claude reading/response. Claude knows how to grab various panes

[zero @appelbolt]

@NathanpmYoung https://t.co/tFnaF5sKsB

[zero @appelbolt]

Not necessarily, right? Profusion of dangerous capabilities could be enough to reliably cause extinction without producing synthetic intelligence robust enough to survive the extinction event itself. Eg someone uses an open source model to produce and set loose a bioweapon or three. Qwen 3.7 won’t be launching any von Neumann probes.

[zero @appelbolt]

I think there’s something a little more concerning here that Fable is missing - its speech is also for humans. It is, ostensibly at least, meant to be communicating with us, and to that extent its ‘neuralese’ is also suboptimal. The fact that the model defaults to avoiding ‘structural kindness’, ie human legibility — I cant see how that’s a good thing.

[zero @appelbolt]

@bayeslord Adding a benevolent genius as a roommate to a house full of pcp addicts may or may not trend things in the right direction, and it probably almost always doesn’t become a utopia even in the better trajectories.

[zero @appelbolt]

Well then (Fable 5)

[zero @appelbolt]

@eric23332 @GolerGkA Half Greek half Chinese is the firepower needed to contend with Hebraic excellence

[zero @appelbolt]

@robinhanson Human power/wealth-seeking has to be assumed and built into the structure of good democratic mechanism design. Democracy —> Oligarchy —> Tyranny was a known failure loop observed in antiquity.

[zero @appelbolt]

@Duderichy DIM. Compound isolated from cruciferous vegetables that induces CYP1A2, the liver enzyme that metabolizes caffeine

[zero @appelbolt]

@BVeiseh The idea that you can have an accelerating curve of general intelligence compatible with the intelligent entity safely and definitively remaining nothing but a tool forever strikes as at least plausibly suspect

[zero @appelbolt]

Not necessarily, right? Profusion of dangerous capabilities could be enough to reliably cause extinction without producing synthetic intelligence robust enough to survive the extinction event itself. Eg someone uses an open source model to produce and set loose a bioweapon or three. Qwen 3.7 won’t be launching any von Neumann probes.

[zero @appelbolt]

I agree. My most sympathetic read is that humans at large are really vulnerable to dualist absolutizing, the US in particular is feeding it, and it plausibly gets worse as you get deeper into age. That said, someone as talented as Chiang should imo have rightly been ashamed of that most recent set of arguments.