Abby Novick Hoskin

@nwilliams030 Arts and crafts room in the Princeton University Art museum. Soooo good. You can also bring some sketch pads and have them draw their favorite art in the exhibit rooms.

If you were taking two 3 year old boys to a little homeschooling field trip at princeton, hypothetically, what places, things etc would you show them? Cool buildings, art, space to run around, places of major scientific discovery, etc!

@thomasfbloom Fair!

@CorpusCalosseum My point was not that one way of solving problems is cheaper/better than the other. (comparing "total cost per problem" is impossible for the reasons you mention.) It was more just that there is a lot of money invested in these solutions, compared to what pure maths is used to.

An aspect of using AI to solve maths problems, rarely discussed, is the monetary cost of running these AIs. For example say an Erdős problem is solved by an AI, and the cost of this run is $10,000. 1/

This analysis seems flawed. It takes a huge amount of money and time to train mathematicians to solve these problems (PhDs take 5+ years). Meanwhile it seems like excellent math aptitudes emerge from general purpose model training runs, so the money you spend training the model can be used to code, do language based tasks, etc. I think it still is “cheaper” to use models to solve these math problems if you account for the cost of training humans + all the other stuff you get from training models.

(As long as the solutions come from their own AI models, of course.) I don't think there's been a time before in history when so much monetary investment has gone into solving problems of this nature. 4/4

@sebkrier Makes sense! Didn't DeepMind sort of dissolve the neuroscience inspired AI team because it wasn't that helpful though? All the researchers I knew on that team have drifted away from this topic.

@CorpusCalosseum Oh I'm definitely not suggesting AI would plateau at brain level capabilities! I just mean the structure could provide some useful heuristics/inspiration for equivalent functions in AI systems etc.

Good read. I do feel like there's still a lot to learn from how brains (and bodies, bioelectricity, cells etc) function, and that this will continue to provide useful insights about both AGI and alignment. asteriskmag.com/issues/13/the-…

@jackclarkSF @viemccoy You can also get a light up table where the light shines through the bottom of the Magna-Tiles and you feel like you're in Star Wars.

@viemccoy this has been the smash hit for the past few months with my kid, and the best feeling in the world is me tapping away on my computer and my kid saying "dada, build with me!" and then we go and make things out of Magna-Tiles.

Becoming a Father has forced me to take a deep look at my digital habits, and ask if I am a good role model for my Son. I imagine that he would be proud to know, even in the womb, that I have just hit 14.2k followers. However, I want him to know the joys of sun and sand, of light and love, pure and detached from the masturbatory media mechanisms we are all so very dependent upon. We must endeavor to strike a balance while we raise him. I fear the neurally entraining beasts of the algorithm, coded tulpas with tentacles leaping out of every display on planet earth, and I will surely spend a great deal of energy to protect him from them. We will need new myths, some of which @calicomccoy has already begun to write, but we will also need new structures and institutions. The old world is no place to raise a child, and the new one is yet to be born (just like my Son).

@deanwball Peak performance

if you have a call with me these days there is a reasonable chance that this is what I look like irl

The majority of women are not interested in sleeping with anyone for a while after they give birth. So the benefits of having multiple sexual partners seem a lot lower for new moms. (Medical advice is to wait ~6 weeks post-birth, but most new moms report it is not fun to have intercourse 6 weeks postpartum.) reddit.com/r/NewParents/c… The cost also seems much higher for women to have multiple male partners around their young kids. Children have a higher risk of abuse/death when they live with unrelated males. (I haven't done a deep dive on this literature, lmk if the economists have debunked it!) uchicagomedicine.org/forefront/news…

New Year’s resolution: have baby #3 🥰

Things get complicated when you have kids. Society is NOT designed for two working parents; once you have multiple little kids (required at societal level for replacement TFR), you need to be making way more than median income to justify not having a stay at home parent. If one parent is going to give up a huge amount of independence/security for the family’s benefit, seems reasonable to demand the protection of monogamy (which doesn’t completely protect a relationship, but seems to be helpful). I don’t think anyone is being abused in this situation.

If you believe AI will produce huge economic effects, shouldn’t you invest your assets accordingly? A lot of people are concentrating on improving prompt engineering, but seems more useful to shift asset allocations? (I am not a financial advisor, I just like it when people bet on their predictions.)

agree with this and @sebkrier's point earlier about the same issue. There isn't all that much concrete I, at least, can recommend people do to "get ready." Step 0 is: internalize what is happening as best you can at both an intellectual and emotional level. Then do as you will.

@sriramk @BrandSanderson Things I accomplished during the pandemic: -Stress baking -Entertaining my baby (not unrelated to the stress baking)

@CorpusCalosseum @BrandSanderson that reveal where he keeps putting books on his table was one of the most insane things ever. (and traumatized pretty much any writer struggling with writer's block)

One of my holiday rituals every year is to go down a rabbit hole of @BrandSanderson content and come away inspired by how awesome of a *human* he is. Brandon is known for his variety of fantasy series set in his Cosmere universe but I originally discovered Brandon's work through an indirect route of wanting to try and write a book of fiction myself years ago. He teaches a course on writing at BYU and the videos for that are some of the best "how to write" content you'll find anywhere. You'll quickly learn a few things about him - the man is insanely productive. He has taken a natural gift and love for writing but then structured his entire life around it. The man can't - he is genuinely kind and generous and his genuine "goodness" shines through. From his interactions with his fans to how he prioritizes his family and faith. In his own words "'m a guy who enjoys his job, loves his family, and is a little obsessive about his stories." Every year I come away thinking the world would be better of with more Brandon Sandersons. P.S For those of you who have never read him, "Mistborn" is probably a great place to start.

@bnzchr @bswud Such a small percentage of people can become/are competent neurosurgeons, it's just blowing my mind the UK pays them so little. And what a bad way to incentivize more people to train up.

@CorpusCalosseum @bswud Watching Americans learn in real time about what happens when a country has a monopsony purchaser of medical labour is one of my favourite things about Twitter.

This neurosurgeon job would be paid an average of $750,000 in the US. We had better hope they don’t make it easy for our doctors to move to the US!

@UnderSecE Love it

@AndyMasley many such cases sadly :( i think people's anger about pits is just due to way more than the fatality stats; parents have to shape their behavior to avoid them. seems crazy when people could just get less dangerous dogs!

@CorpusCalosseum Ugh I'm so sorry that's terrible, didn't know

I see this graph of pit bulls as especially dangerous come up a lot, and I feel like people are thinking way too much about the relative proportions and not enough about absolutes here. 284 deaths between 2005-2017 is still relatively small. To bring it back to my AI enviro stuff, this feels somewhat equivalent to saying "ChatGPT is 10 times as bad as a Google search." That's still just so small. Here, it's saying "Pit bulls are 20 times as dangerous as huskies!" Like okay, when was the last time you encountered any danger from a husky? An extremely small number x20 can still be extremely small. Seems like there might be ~10 million pit bulls in the US. Of like ~10 million dogs, every year ~23 people are killed. 1 death per 430,000 dogs each year. To compare, there are ~284 million personal cars in the US, and 40,000 road deaths each year. 1 death per 7100. So the average car is 60x as likely to kill you as the average pit bull. I'm obviously pro being very safe with very physically strong animals. But I think a lot of people infer way more from this graph than they should. This seems kind of like a graph of the least safe airlines.

I thought my 5 year old was thankful for @awscloud, but it turns out she just likes having arms. And retweets.

Went to the @JoinFAI Gala and all I got was a nice photo of my dress and a deranged photo of me fangirling a Nobel laureate