My kid was asking me about the likelihood of aliens in the universe and got all freaked out, yet thinks AI is cool and funny. Made me think we should start referring to it as Alien Intelligence and memeing the idea that we’re building an Alien, not something Artificial. Seems more resonant with normies and understanding the stakes. Also arguably more accurate.
I think the "Do people think AIs are sentient?" survey is being confounded with the more fundamental question of "Do people understand what 'sentient' means?"
"Imagine asking someone for advice if all your conversations had to be posted online as transcripts, and how that would change your behavior, now understand that many people think that would be good."
This seems like applying the Most Forbidden Technique to humans, with similarly undesired results...
Didn't "mirror neurons" get debunked/fail to pass replication? Coulda sworn Scott discussed this at some point...
The SOO paper makes me wish for more funding in alignment, better researcher access to frontier models, prerelease testing, etc...it's a neat demonstration! But only on weaker models that are cheap to do science with. "Works on Mistral" is like the AI equivalent of "IN MICE". Except even more than cancer cures or whatever, here we care almost entirely about what happens on the long tail of frontier models only, so generalizing from a weak case ranges from less-useful-than-expected to actively-misleading. Also illustrates the frustration with open source - clearly very useful for research, but the higher those capabilities get...
The AI memes are...I mean they're not not memes, clearly some of the fundamentals are grasped. But it's like going through an extra interprative step where something gets lost in translation. Partly in the formatting (It's The Principle Of The Thing, consistency is important for expectations), partly in the "you're using Setup X with Meme Image Y" thing where they don't quite match up. For now I think we're still in the Impressive Slop phase, while admitting of course that quantity has a quality all its own. Some of the more linguistic-based ones might work better? The recursive nature of "yo dawg, we heard you like yo-yoing, so we put your dog in a yo-yo..." seems ripe for AI innovation, for example.
Relatedly, that's also why I disagree that people won't want to "connect" with "AI creatives". Trying to understand the alien thought process behind subtly-off humour is *more* interesting to me than figuring out what makes Louis C.K. tick. We understand the fundamentals of human humour pretty well, the rest is just in the execution. It's like picking up the humour from non-Western cultures, suddenly you've got a whole new area of mindspace to explore and it's got that shiny scent of novelty that rereading Samuel Johnson on why puns are outre.
Anthropic appear to be advertising a mid to senior developer position in London for 255k-450k GBP
Not sure what to make of that. It’s so far above local market rate as to be ludicrous. (You could get one of the best senior developers in the UK for a quarter of that top number. 450k gets you god.)
Is it fake? Has anyone actually been hired on these numbers? Should I immediately quit my technical leadership position to apply for this mid-level developer role?
Why wouldn't it be real? I mean, isn't that exactly the reaction they want? Isn't it a good deal for them? You code just as well in London as others do in SF.
I've been an improv performer/instructor for 15 years. By volume most improv shows are pretty bad, because the shows themselves are consumption goods for the performers who are typically the students in improv classes. I haven't ever tried doing a scene with claude, I'm now curious. I'll do one and if people are interested give my review of claude as an improv scene partner
Podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/ai-108-straight-line-on-a-graph?r=67y1h&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
The very top graph doesn't show up here.
> I wake up
> another straight line on a graph
Good thing my knowledge and abilities double every 7 months or this would be bad
My kid was asking me about the likelihood of aliens in the universe and got all freaked out, yet thinks AI is cool and funny. Made me think we should start referring to it as Alien Intelligence and memeing the idea that we’re building an Alien, not something Artificial. Seems more resonant with normies and understanding the stakes. Also arguably more accurate.
I still like invasive species. Doesn’t imply sentience
Sure, but I think renaming AI is a snappy memetic way to spread the idea of inherent risk
Artificial Invaders!
I think the "Do people think AIs are sentient?" survey is being confounded with the more fundamental question of "Do people understand what 'sentient' means?"
"Imagine asking someone for advice if all your conversations had to be posted online as transcripts, and how that would change your behavior, now understand that many people think that would be good."
This seems like applying the Most Forbidden Technique to humans, with similarly undesired results...
Didn't "mirror neurons" get debunked/fail to pass replication? Coulda sworn Scott discussed this at some point...
The SOO paper makes me wish for more funding in alignment, better researcher access to frontier models, prerelease testing, etc...it's a neat demonstration! But only on weaker models that are cheap to do science with. "Works on Mistral" is like the AI equivalent of "IN MICE". Except even more than cancer cures or whatever, here we care almost entirely about what happens on the long tail of frontier models only, so generalizing from a weak case ranges from less-useful-than-expected to actively-misleading. Also illustrates the frustration with open source - clearly very useful for research, but the higher those capabilities get...
The AI memes are...I mean they're not not memes, clearly some of the fundamentals are grasped. But it's like going through an extra interprative step where something gets lost in translation. Partly in the formatting (It's The Principle Of The Thing, consistency is important for expectations), partly in the "you're using Setup X with Meme Image Y" thing where they don't quite match up. For now I think we're still in the Impressive Slop phase, while admitting of course that quantity has a quality all its own. Some of the more linguistic-based ones might work better? The recursive nature of "yo dawg, we heard you like yo-yoing, so we put your dog in a yo-yo..." seems ripe for AI innovation, for example.
Relatedly, that's also why I disagree that people won't want to "connect" with "AI creatives". Trying to understand the alien thought process behind subtly-off humour is *more* interesting to me than figuring out what makes Louis C.K. tick. We understand the fundamentals of human humour pretty well, the rest is just in the execution. It's like picking up the humour from non-Western cultures, suddenly you've got a whole new area of mindspace to explore and it's got that shiny scent of novelty that rereading Samuel Johnson on why puns are outre.
What is the ‘can’t silently drop certain sources no matter what’ rule?
What do people use 4B and 1B models for?
Anthropic appear to be advertising a mid to senior developer position in London for 255k-450k GBP
Not sure what to make of that. It’s so far above local market rate as to be ludicrous. (You could get one of the best senior developers in the UK for a quarter of that top number. 450k gets you god.)
Is it fake? Has anyone actually been hired on these numbers? Should I immediately quit my technical leadership position to apply for this mid-level developer role?
Why wouldn't it be real? I mean, isn't that exactly the reaction they want? Isn't it a good deal for them? You code just as well in London as others do in SF.
Because companies don’t usually pay hundreds of thousands of pounds more than they need to!
Could be they’re committed to a single global payscale for cultural reasons?
But yeah they’ll have their absolute pick at that number
I've been an improv performer/instructor for 15 years. By volume most improv shows are pretty bad, because the shows themselves are consumption goods for the performers who are typically the students in improv classes. I haven't ever tried doing a scene with claude, I'm now curious. I'll do one and if people are interested give my review of claude as an improv scene partner