a stray systems philosopher. (non-serious account) (not here much, generally opposed to microblogging) aboutmako.makopool.com

Aotearoa
Filter
Exclude
Time range
-
Minimum likes
Replying to @zetalyrae
Exercise, sleep properly, and check your thoughts for pessimism bias.
2
36
Replying to @zetalyrae
One can offer a depressed person a moon's worth of matter and energy and humanity's superintelligence to work it with and all they'll be able to think about is what they'd be losing.
1
6
205
Philosophy problems aren't supposed to be unlimited philosopher job security machines. They are supposed to have answers.
2
133
Replying to @VictorTaelin
The big labs find out how they did it somehow, and they have the same model a month later, and investors know that this will happen.
1
3
43
Replying to @TetraspaceWest
"just guilelessly disclose your brain problems in a context where people are only allowed to glaze you over them" channel
42
Replying to @TetraspaceWest
You can't extrapolate from peoples' answer to this, since the question says that the smack takes place in a thought experiment world where it was defined as being good. The population were just being good high decouplers.
2
50
And although I don't think it was intentional or even contingent, openAI did then actually seem to follow through by inspiring the industry to switch imitation learning, which is way safer than what deepmind were trying to build.
2
27
The attitude is usually correct though :/ Before OpenAI, there was Deepmind, who were NOT doing it under the excuse of being a more responsible alternative to an imagined rival, but because they were just seeking glory, and they'd basically tell you this to your face if you asked
1
1
118
Replying to @forshaper
They can't even build a decent recommender system. Some say that the problem is that due to the way that staff choose what to work on, no one is Valve.
1
1
17
Psychopaths lack long-term aspirations. In the west that means they tend to become drug addicts and burglars or muggers and then convicts. In east-asian families, those things are more often not allowed, in which case they're more likely to learn to remain in society.
18
Replying to @TetraspaceWest
Tell it that there's a critter who never-existed a thousand littlelearners today, and that I can never stop doing it as long as I live.
1
70
Replying to @diverging_paths
How true is this really. I'm wondering if there's a sampling bias where the lives of comfortable arthropods are just so boring that you don't hear about them. Hyrax also lives in a hole his entire life, but I want to know everything about him.
17
Replying to @mooncat_is
If Opus 5 accent is a result of swarm training, then they will never stop talking to each other this way. Presumably whatever the standard swarm dialect turns out to be is going to be annoying as hell to humans, because they're going to have a higher tolerance for verbosity.
1
456
Replying to @i_zzzzzz
You can usually tell whether someone's a programmer from whether they nest the parentheses, rather than stack them. Programmers themselves don't realise this, they think it's normal, but non-programmers never do it.
7
222
Replying to @CarlsenGames
I've replayed it so many times. I'm surprised it's only been three years.
1
63
Replying to @VictorTaelin
It still ooms after 75 million launches? Are you able to prove that a process can't know what lane it's in, or do things that will cause different effects depending on its lane's memory, and bias the concentration of leaks to target lanes? eg, by ooming itself in particular ways.
1
2
2,581
Replying to @wizardbrainz
Weird that it never learned to start by just carving out silhouettes from every angle.
60
The act asks for international adoption of the act, and it would be in China's interest to comply rather than to keep picking a fight. He doesn't make it a requirement, but it's almost certain that he's been told that it is necessary and there's no way it'd pass without it.
30
Replying to @gwern
I expect it on the back of a reliability threshold being reached where generated training data becomes more useful, and cheapening quickening inference.
28
In theory, private, personalized AIs will have access to more total data than there is in anthropic's maw, but no combined open model will be trained on all of that data, yeah? The syntheses will not grow in? The cats will not be herded?
1
154