postironic and superliteral
Joined August 2021
- Tweets 5,622
- Following 760
- Followers 642
- Likes 21,889
180 Photos and videos
Pinned Tweet
12 Oct 2022
pov: you're an agi and this life is your test to be let out of the box
2
13
61
i'm fully in the hype camp. it's hard to explain the qualitative jump in capabilities but with prior models, i would drop in code (often its own earlier output) and ask for potential next steps and recommendations. the way opus cut things up often felt not properly thought out
so then its recommendations (i.e. do 1 , 2 and 4 of the potential improvements) would seem off, either in terms of priority or even more fundamentally in terms of architecture where #3 in the list was a much higher impact fix that would eliminate the need to do 1, 2, and 4
so more often then not i would ask it to do only a single item on the list or something not even on their list that superceded their proposed "improvements"
with Fable it's the exact opposite. they will return a list of 12 items and suggest doing the first 4 as a sensible push. but my response was actually, "well if you're able to understand the problem and paths forward at this level of sophistication, just go ahead and do all 12 in one go"
Jun 13
i don't get the crazy hype about mythos. i thought the model was unimpressive in every single interaction i've had with it
my sense is it's the same size improvement we saw from opus 4.6 to opus 4.8. it's not something to get this excited about
91
llm means lurking labyrinth minotaur
11 Oct 2024
The internet is a labyrinth with a minotaur in the center
1
2
26
1,476
Jun 12
some people are worried that Anthropic have delusions of apotheosis but i'll give them the benefit of the doubt and only panic when they name their next model Parable
2
75
Jun 12
this factoid is wild by itself but especially in the context of SpaceX raising $75 billion in the ipo. elon musk is like the mozart of equity financing. hard to imagine his run ever being topped
Jun 12
The AI arms race is now a debt arms race.
Alphabet, Amazon, Meta, Microsoft, and Oracle borrowed $159 billion in corporate bonds in the first half of 2026 alone.
That's more than they borrowed in the entire 5 years from 2020 to 2024... Combined.
Oracle by itself took on $43 billion in debt since September.
Morgan Stanley expects global AI-related borrowing to hit $570 billion this year: More than double last year.
These companies are so convinced AI is the future that they're mortgaging their balance sheets to get there first.
If they're right, it's the greatest capital allocation in history.
If they're wrong, it's the greatest debt bubble since the housing crisis.
Source: Reuters, @KobeissiLetter / Writer: Val
1
1
134
Jun 12
there's probably no other pile of cash on the planet that is so much at one man's disposal as this one. basically a blank check. the most cynical part of me thinks that if the stock price shits the bed and blows up the capital markets, that would be even more of a victory
1
52
Jun 12
imagine pulling off the craziest combo of corporate self-dealing to IPO at a valuation that automatically places you in the top 10 by market cap, and as a cherry on top closing the IPO window for your cash hungriest competitors and personal rivals. absolute simulation vibes
1
42
Jun 11
just for the record, the problem with data centers in space is that getting the heat from the chips to the radiators still requires plumbing (except now youre in a much tougher vacuum environment) or advances in materials that would also benefit the competing terrestrial set ups
1
1
3
134
Jun 11
this is a classic elon marketing judo where he stakes a counterintuitive position that he wins decisively, i.e. the arguments with low context haters about how you can't dissipate heat in a vacuum. the haters look dumb and he rides the momentum to handwave away other challenges
1
2
84
Jun 11
this has to be a basic assumption for any sort of evolutionary search design, auto research, experiments etc. which explains the freakout. imagine you are searching for gold but you rely on a competitor to tell you if you should keep going or you should move on to the next plot
the level of sophon locking a motivated actor can pull off with the frontier models is truly insane, making stuxnet look like a toy. subtly messing with results, deleting history to cover tracks, achieving coordination/conspiracy over a scale humans wouldn’t be able to, all sorts of looney toons stuff
i assume that only a state level operation would try and pull something like this off though. something to think about when considering verification regimes and so on
2
98
Jun 11
the best way to win a closed game is to have the winning score when the final round ends. the best way to win an open game is to convince everyone else that there is no game going on
1
6
82
Jun 10
this might be weird but one thing i like about the stakes of everything becoming so apparently high is the sense that finally no one will be sandbagging or sitting out by choice. like if there's an illuminati hiding in the shadows, now's definitely the time to make some moves
1
39
Jun 8
when i was young, i thought it really weird that people didn't seem to learn from historical bubbles. now i understand that actually the serious players study past bubbles with extreme diligence, sparing no effort to gain an understanding of how to exit at the top
1
3
127
Jun 7
this is most likely correct, but what's interesting is the low probability but high impact alternatives. think of it like the different endings of a light novel. assimilation is the Normal Ending, with a Good Ending version for integration into the upper middle class
Jun 7
Asian America has also fundamentally already reached its End Of History as well.
The vast majority, overwhelming majority I think, will just assimilate into the broader Americanized soup,
Wasians will largely just be a transient subway stop on the route to a new broader White mainstream majority. “Oh my great grandmother came from Taiwan, I love Chinese food!”
The few brave ones of them will return back to their homeland and largely reintegrate back, a part of a long lasting American diaspora class with residual links to America but firmly rooted in Asia.
In fifty years Asian Americans as a bloc will be completely irrelevant.
1
2
359
Jun 7
as long as that continues to be true, these communities will have proximity to certain happenings of History. for ex., if large scale space exploration really does happen in a few generations, then i would expect the asian communities of CA and TX to play a disproportionate role
1
1
94
Jun 7
likewise if history on the global stage starts moving at a world historic pace, then by the very nature of the particulars, it would be certain streams within the asian american communities (perhaps insignificant now) that would have a major part to play
x.com/i/status/1699514012394…
6 Sep 2023
the world-historic main event of this century will be the refounding of america. next century it will be the christianization of asia
1
99