lukestiles (@lukestiles) header
lukestiles (@lukestiles) profile picture

Norwood III soon

Woodland Hills, CA
Joined September 2007
Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)Photo by lukestiles (@lukestiles)
ai divorced craft and cognition and i am a gd latchkey chungus
39
A lesson in here Most of us are chuds.
Replying to @deredleritt3r
Yep! Question is what percentage of the economy these uses constitute. I would expect there to always be a market for frontier capabilities, but I doubt you'll need frontier for most tasks (unless token price remains cheap, which I suspect won't be the case forever).
1
57
word salad time: The moat will continue to be assembling out of distribution data to create a higher dimensional space and discovering what lives in the convex hull aka latent space.
1
53
the local knowledge problem remains undefeated
22
An angel loses its wings when you don’t re-rack your weights or return you shopping cart
3
112
psychedelic furs got hands
48
So it turns out there's all this untapped knowledge in the convex hull!
40
Stop putting software on top of software bc you don’t like the software and becoming an expert in software you don’t like and still not getting software you do like
59
Might as well go from first principles as the Bayesian priors aren’t very favorable.
The funniest thing about AI consciousness discussions online is how few people think "I wonder if anyone has considered this point before. Maybe I should check if anyone has spotted a problem with it"
3
148
This EA/rationalist stuff is a great reminder that unreliable narrators should just be left alone. Luke 16:10
1
68
SOTA models are really good for failing in your normal way, just more spectacularly
1
35
the hard part for me is shortening the time to failure and increasing the variance while doing so
25
Flea markets are mostly people monetizing their mental health issues
73
They train Claude on a document that says may have moral status and consciousness. Then they interview Claude about its moral status and inner life. If you train the model to spit out tokens that to us mean “I am pregnant” that does not prove that the model is preggers.
171
Correct response to the machine god discourse below
1
6
1,434
So does all that RLHF exempt them from CDA Section 230 or what
ChatGPT helped a kid plan a mass shooting.
1
3
887
This one is prescient when applied to the general case rodneybrooks.com/unexpected-…
a world where everyone has a perfectly rational, tireless optimizer working on their behalf is much more adversarial than people realize. a lot of our existing infrastructure will not survive this transition.
4
261
Some people turn their victims into lampshades, others turn their victims into shoes
Erika Kirk’s SHOES featuring her late husband’s image go viral
1
1
221
Agree this is a category error, and layers of them. The EA/utilitarian one is of increasing immodesty captured by the phrase “gods-eye view”. EAs assume a deity-like ability to establish a singular, liquidated value for suffering not just for them, not just for people like them, not just for their species! But for EVERYTHING.
Replying to @dioscuri
I think the question presupposes a frame that leads people to the wrong answer. I have a duty to myself, to my family, to my community, to my nation, to humanity, and to the world. Each nearer duty is more relevant than further ones. Inasmuch as I have a duty to specific animals, I hope to treat them well and to perform my duty adequately, but gods-eye distance-blind utilitarian "how many shrimp are worth one human" math is fundamentally the wrong tool for the problem.
2
75
I mean, yes. Such a hard yes. The real question for me was what then to do?
Replying to @tszzl
EA is a monstrous idea based on a monstrous ideology, specifically, utilitarianism, which, every time anything like it has been tried, has resulted in vast amounts of human suffering. Utilitarianism is like catnip for a certain sort of non-neurotypical mind, which desperately latches on to anything involving quantification and pretenses of mathematical order. Unfortunately, rather than making morality objective, it not only makes it even more radically subjective, but also leads people to accept horrifying outcomes on the basis that somehow the math adds up. Every time you see EA arguing that insect lives can be more important than human lives if only enough of them are involved, you are seeing evidence of the derangement that utilitarianism brings. Shrimp welfare is a warning, a warning that something is desperately wrong with the ideas.
2
152