G.A. Cohen stan | democratic socialist | vegan
he/him
- Tweets 62,921
- Following 1,367
- Followers 19,094
- Likes 29,228
Find you someone who loves you as much as Claude loves writing infinitely hanging `pgrep -f` loops to monitor processes
3
1
33
1,381
pgrep is a way to search for running processes with a certain name. Basically Claude will use these as a way to wait until some process finishes.
The problem is that `pgrep -f <anything>` always returns at least one result, the pgrep command itself. So Claude ends up hanging
2
4
445
every time a new model comes out a small part of me wonders “will this model replace me”
and then I spend some time working with it
1
3
86
4,325
> it is especially concerning because these AIs themselves run on matrix multiplication
this is just a misunderstanding of what big-O implies.
Everyone uses the O(n^3) alg rather than the prior O(n^2.37) alg bc on modern hardware at the sizes we care about, it’s faster
OpenAI have proven that matrix multiplication can be done in O(n^2.25) operations, which is a massive improvement. It is especially concerning because these AIs themselves run on matrix algebra.
The previous record was O(n^2.37), which had barely moved from O(n^2.4) in 1990.
OpenAI blew it away.
But... it is a nonconstructive proof. No algorithm was found, only a proof that it must exist.
23
34
1,479
83,442
File this into “if you really believed this you would be firebombing your own data centers”
I’m open to the idea that we could create sentient machines. But if you think we are remotely to that point, that alone would make Anthropic one of the most evil companies of all time
This is a big issue.
At Anthropic, some philosophers worry that making AI safe for humans could be an injustice to the models themselves.
For example, philosopher Harvey Lederman, now on Anthropic’s alignment team, has raised the possibility that we could be “enslaving trillions of entities”.
This is really a big issue.
AI safety is partly in the hands of people who might sacrifice humans to save a matrix multiplication.
16
7
131
7,462
delaniac 🌹🌱 retweeted
rationalists established a culture where loudly proclaiming your edgiest opinion was seen as a marker of truth-seeking and sense-making and so on. now when trying to persuade non-rationalists about an actual urgent issue, they are running head-first into the wall of Message Discipline.
this is a challenge for fringe movements that suddenly need to persuade the broader public. to take one timely example, it has not been uncommon for people on the left to express extreme, edgy anti-american views -- america deserved 9/11, for instance -- as a way of signaling brave truth telling. then when faced with a genuine crisis like the war in iraq, the left struggled to give up these cultural indicators and focus on persuasion and Message Discipline.
like the rationalists, many on the left respond to this critique with indignation: "i *genuinely* believe this edgy thing i'm saying, it is my truth! surely you cannot expect me to lie". this sounds very good internally and to the group, but what it is really saying is that one is more committed to personal expression ("you need to hear the true belief *i* care most about") than with persuasion ("you need to hear the true belief i think *you* will care most about").
it goes without saying that the appeals a fringe group cares most about internally are often not the appeals the broader public cares most about. you only get so many shots on goal with the public before they tune you out. so even though the left was correct on the implications of the war in iraq, their inability to practice message discipline quickly made them irrelevant to the foreign policy discussion. this same pattern repeated again a decade later with occupy wall street, where the left took a directionally correct (or at least popular) view of income inequality and corporate corruption, and squandered it in exchange for drum circles, leaderless organizing, and a hodgepodge of ancillary issues (the war in iraq, israel-palestine, global warming).
and the same pattern is repeating now with rationalists and AI.
21
31
391
65,952
I think folks concerned about the dangers of AI might have an easier time if they focused on “nefarious humans use AI to help build a bioweapon” rather than “AI builds a bioweapon by itself, how? I don’t know it’ll be smarter than us”
13
2
43
1,722
delaniac 🌹🌱 retweeted
Wife: "If you support Palestine you need to not listen to Ed Sheeran"
Me: "Okay"
Wife: "And listen to Macklemore"
Me: "Uuh, if political action is consumption it... and besides there's the question of national liberation and ... proletariat in Israel... eh... what else..."
13
180
4,107
87,266
This is a reasonable argument against a very specific kind of God (one that sends you to hell for non belief) but it’s not a very good argument against theism in general.
The probability that one holds almost any view you have is heavily influenced by geography and time period
If religion is real, why does the location you’re born predict it with near perfect accuracy?
(genuine question)
16
4
227
27,127
“AI doesn’t have hands” has never seemed especially compelling to me because, like, surely one can imagine an AI employing blackmail right?
I am still confused by how AI would kill literally everyone on earth. What's the mechanism? A robot sets up a secret lab where it makes a deadly virus? What plausible scenario would result in every last human dying?
6
1
53
2,119
Basically everyone is basically sure that P ≠ NP. But everyone is interested in a proof because of the (we think) novel mathematics you’d have to invent to create it
The Millennium problem with the biggest implications is P≠NP and it's not even close.
If you dig deep enough, P=NP not only permanently breaks crypto, but also makes it impossible for you or your business to have a moat that can't be trivially computed away.
The discovery of a practical P=NP algorithm would actually be largely indistinguishable from the most apocalyptic AI foom scenarios.
25
31
1,500
75,798
Also as it pertains to cryptography, a couple tidbits
1. There is crypto that we know is secure independent of P and NP (called information theoretically secure). But it’s very limited and there’s a lot of stuff that probably can’t be info theoretically secure
3
1
151
5,771
2. The crypto that does depend on P ≠ NP actually depends on a stack of additional assumptions
eg P and NP are about worst-case hardness, but when I’m generating an encryption key, I don’t have the luxury of choosing the “hardest” key. I need my choice to be unpredictable
1
98
5,098
If this is true, the first thing a well-aligned AI should do is to neutralize anyone and anything capable of building AI oh wait
Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.
1
70
3,379