“What must it be like to live in this world, seeing it just the way it is, and think that it will never change, never get any better?”
Washington, DC
- Tweets 53,915
- Following 5,486
- Followers 8,275
- Likes 428,784
Mad respect to my ancestor who died at the Alamo, but honestly the Texas Independence movement was one of the least justified independence movements in history
The Texans basically did a hostile takeover of Texas shortly after immigrating there
2
24
595
The thing I like about Dario Amodei and Bill Gates is that they are very prickly in interviews
There’s something endearing about seeing someone argue their beliefs without trying to pander or win a debate competition on points
6
291
I feel like “don’t let your super-cyberhacker agents have indirect access to the internet to install Python packages” was too obvious for the AI doomers to be yelling about in 2019
if people have been worried about loss of control and containment for this long, it feels kind of obvious that there should have been teams in the ai safety community spending a lot more money on the practical work, like building strong sandboxes, trying to break them, formally verifying, etc.
wondering, why wasn’t this a much bigger part of ai safety work from long ago?
3
1
16
783
The thing about making predictions early is that I can brag if I’m right, and nobody remembers them if I’m wrong
My current baseline scenario for the midterms is that Trump manages to end the war in the next 3 months and he settles into being a lame duck
Trump’s approval rating shifts back to 42% and the midterms are D+7 with Democrats narrowly taking the House and getting 49 Senate seats
3
41
1,747
I’ve listened to a lot of what Dario says, and he seems more grounded and skeptical on the idea of AI consciousness/“soul”
None of the AGI concerns about reward hacking, cybersecurity, bioscience, surveillance, etc. require you to believe in AI consciousness
I genuinely don't get Anthropic. It seems like every other day they're loudly puzzling over the ethics of a soul in a box - while building the box, claiming that it has a soul in it, and selling access to said box to for $20 a pop.
Like, dudes, these are incompatible positions.
2
11
418
For people who are concerned about making AI art/literature/movies without stealing someone’s IP, it might be possible within the next decade to pre-train an AI model with uncopyrighted (or creative commons licensed) works and then fine-tune it on your own intellectual property
This seemed had to believe at first (given that a studio might have the rights to, say, 1000 films?)
But ChatGPT thinks that you’d only need ~200 hours of film to reasonably fine-tune an open source video model
So this fine-tuning might actually be feasible
1
7
609
ChatGPT made my tweets better btw. I was originally going to quote tweet with skepticism
But instead I asked ChatGPT about whether this sounds reasonable
This seemed had to believe at first (given that a studio might have the rights to, say, 1000 films?)
But ChatGPT thinks that you’d only need ~200 hours of film to reasonably fine-tune an open source video model
So this fine-tuning might actually be feasible
5
569
