Glorbutron ⏹️ retweeted
Wow, every action you've taken for the last month has involved strategically avoiding doing the one thing that you believe is most important towards achieving your goals. That's a pretty smart way to avoid risk and disappointment
Glorbutron ⏹️ retweeted
my spidey senses are telling me the labs are, as we speak, using unguardrailed versions of their unreleased models to systematically reverse-engineer ALL of the internet’s software, apps, and games, then feeding the reconstructed source code straight into their training pipelines while simultaneously building guardrails to ladder-pull the rest of us
Glorbutron ⏹️ retweeted
at all times, around 25% of the total foveal surface area of all humans who are awake and looking at anything is covered by light from pixels. literal imprisonment of the human spirit in a flat world of false light ordered by the most cynical incentives you could possibly imagine
I strongly disagree with 3. In much the same way as it is possible to produce guarantees about control systems I expect similar guarantees about things like stability under reflection or limited self modification will be feasible.
I agree with Scott Alexander's recent article on AI safety. I add three points:
1. We are not in existential danger right now.
2. Being an enemy of the AI companies makes you less credible.
3. We will never know, definitively, that we've solved alignment.
1
160
Glorbutron ⏹️ retweeted
Glorbutron ⏹️ retweeted
I think most of the hand-wringing about life losing some important qualities post-singularity is misguided, because if we do (somehow) get perfectly CEV-aligned superintelligence, it will simply solve all these problems
if we don't, you have much bigger things to worry about
Glorbutron ⏹️ retweeted
let’s not forget that Youtube figured out how to make internet comments helpful and kind and every other social media platform decided not to follow suit because they believe outrage and division is better for their bottom line
teapawt.org/RichDecibels/status/17…
Glorbutron ⏹️ retweeted
Guy Debord argued that modern life was increasingly being replaced by representations of itself. In 1994, he died by suicide. David Foster Wallace proposed sincerity as an escape from recursive irony. In 2008, he died by suicide. Mark Fisher wrote about the ghosts of futures that never arrived. In 2017, he died by suicide.
Now it is our turn to solve postmodernism.
Glorbutron ⏹️ retweeted
New Anthropic research: How AI agents shape experts' judgment calls
When experts delegate work to an agent, do they still make the key decisions? We tested this in a randomized trial with 50 professional data analysts.
in the past i never really bought all the arguments around not allowing online voting, and generally requiring physical ballots and airgapped machines. i thought figuring out good cryptographic digital solutions was critical to democratic participation.
holy fuck was i wrong. so incredibly glad we've kept this line as a society, looks amazing prescient. extremely happy we won't go into 2028 with massive cyberrisk hanging above our elections.
The Copenhagen Interpretation of Ethics arises as a natural defensive instinct amidst moralities that fail to treat aware inaction as culpable as aware action.
One reason festivals, for example, don’t want to know how much rape happens there is that organizers worry they will be held legally responsible if a lawyer can show that they knew. Not caring is safer.
1
3
108
Glorbutron ⏹️ retweeted
I've been trying to figure out why coding with AI is so miserable. It sounds like it should be fun, like having a magic wand that brings your dreams to life.
In my case, I think it's mostly because, for the sake of speed, I've decided not to try understanding the code myself. While this is extremely effective in the short term, it becomes extremely stressful as the project grows in complexity, because you've rendered yourself completely dependent on the AI for any kind progress. If you ask the world's smartest AI to try to fix a bug and it fails, it feels nightmarish, like you're completely trapped and there's no way out. Your only options are to impotently keep begging "pls fix," or to try and fix it yourself, but by this point, the codebase is a spaghettified monstrosity.
And so the unpleasantness is a consequence of intentionally disempowering yourself.



































