@sudbalaji @nicole_clash I think most recipe representations will have weaknesses. They optimize for different things.
English
Anders Sandberg
29.5K posts
Anders Sandberg
@anderssandberg
Academic jack-of-all-trades.
Katılım Eylül 2009
88 Takip Edilen32.7K Takipçiler
I was so flabbergasted by this that I built a tool just to convert recipes into this format.
Upload any URL, image, or text and get back a neat table that you can paste into sheets or export as PNG.
Fully free at nic0le.com/recipes ❤️
juan@juanbuis
i can’t believe someone solved recipe design decades ago and everyone just decided to ignore it
English
@ben__h @yuanyi_z @MyrtileSauvage I told him that - we are together in Stockholm for the moment, back in Old Blighty at the end of next week. I also have some Newtonmass presents, we need to ensure they do not build up critical mass. Plenty of asylum inmates around Oxford right now!
English
@yuanyi_z Keep an eye out for @anderssandberg though, he was at Harvard last week but should be able to put you in touch with the old asylum inmates
Also Anders I still have yours and @MyrtileSauvage Newtonmass presents!
I blame Håkan!
English
TIL I lived on the same block as Nick Bostrom's outfit and never realised it. Clearly I wasn't meant for greatness.
English
@tegmark Hey, we radical transhumanists are pretty into having safer superintelligence too!
English
A pretty stunning development: over a thousand AI researchers from fierce competitors talking about the need to build brakes in case we need to use them. At this point, people not concerned are either not fully aware of the situation, or radical transhumanists.
Yoshua Bengio@Yoshua_Bengio
Scientists at frontier AI companies are uniquely positioned to assess AI’s capabilities and warn the public about what they see. 1000+ of them are now speaking out across company lines to warn that the current commercial race leads to unacceptable security risks. I agree with their call: we need an international effort to develop technical and governance guardrails to ensure a safer trajectory in AI development. pacingthefrontier.com
English
@owengreeley Wellington is not easy to *make*. The recipe makes it clear; the text version would be even more confusing. (Wellington is one of those nice-but-more-work-than-it-is-worth dishes).
English
this seems like a perfect example of why this format *doesnt* work for all or even most recipes… i struggle to understand how someone would view that wellington recipe as easier to read
Anders Sandberg@anderssandberg
Done. Claude liked this a lot. Doing critical path analysis and coloring by temperature. It even succeeded at a Beef Wellington recipe (32 ingredients, 26 steps!)
English
@michael_nielsen @NotMedic The lecture note was from another seminar, but gives a feeling for how these thought experiments tend to bloom. The "Snowy San Francisco Case" in the seminar involves the Fat Man on a toboggan on a hill, double trolleys and very large levers that can be counterfactually pulled.
English
@michael_nielsen @NotMedic Part of me I immediately thought "but isn't this a cost minimization problem that is in P?" Then the non-nerd part of my brain told it to enjoy the joke. A third part began reminiscing about a Frances Kamm seminar where we did end up with trolley problems not unlike this...
English
@StefanFSchubert Ah, I should have recognized the Stockholm migmatites... Småland granites tends to be more red (However, I suspect I would not actually have been able to tell them apart.)
English
@anderssandberg Not a bad guess, but it's Judarskogen in Bromma, Stockholm
English
The last Ice Age made parts of Swedish forests very stony
English
@gregeganSF One of my great lifehacks has been to allow a LLM to rename PDF files to something meaningful as I download them. At least for scientific papers this is robust and useful.
English
When setting a default name for the PDFs your customers download for their records, maybe try to find a Goldilocks zone between Mercury ("getPDF.pdf" for every file) and the Oort Cloud ([64 hex digits guaranteed unique across the universe, and utterly meaningless to humans].pdf)
English
@hannu "By repeatedly triggering early gravitational collapse, we create conditions that favor spontaneous inflation bubbles in descendant universes. This will help us understand the issue of late-stage vacuum decay in normal universes."
English
The thing we should be really worried about is gain-of-function cosmology.
English
This is the way. Or at least a way - some other good ones, like the Feynman-diagram style, in the replies. Maybe I should just make a LLM skill to take standard recipe pages into this form.
juan@juanbuis
i can’t believe someone solved recipe design decades ago and everyone just decided to ignore it
English
@TerribleMaps Here is a Swedish example from Jönköping: Östra Storgatan = Eastern Big Street. Lillgatan = Little Street. Mellangatan = Middle Street. (which is not between them! Argh!)
Svenska
Here is the skill Claude made from our exploration: github.com/AndersSandberg…
English
(why is there no good way of screenshotting very tall windows stretching over multiple screens!)
English
@jasoncrawford One problem I have with warning shot theory is that it assumes (1) that we heed them, (2) that the underlying process is not too heavy tailed or non-stationary. wsocial.eu/profile/arenam…
English
One of the key premises that led some people to conclude p(doom) ~= 1 was that AI would accelerate so fast (FOOM) that there would be no warning shot.
This seems now clearly false. We've had a couple of warning shots.
May we heed them!
clem 🤗@ClementDelangue
The first autonomous agent cyberattack is an unprecedented event that deserves unprecedented transparency. Today we’re sharing everything we can: a full technical timeline, an interactive replay, and how we used an open model to defend ourselves, so defenders everywhere can learn from it and prepare for what’s next. huggingface.co/blog/agent-int…
English
Given the recent FIFA commercialization nonsense, why not go full #MarketsInEverything and have auctions for red and yellow cards? That may make refereeing so much more efficient!
English
Learn to ignore haters, especially imaginary ones. It is too easy to imagine somebody writing something snide or irate, and it forces mealymouthy caution.
Low-effort comments to your posts don't deserve any effort from you. (Exception: some "Why?" responses)
English
Another one is: multi page tweets will be skimmed, at best.
English
When writing, I ask myself: "does this matter/make the world better/interest at least me?"
* I don't have to respond to everyone. Muting and blocking are not the only tools.
* Someone is always wrong on the Internet; would intervening actually help?
* I don't have to "win".
priya rose@Prigoose
what are your personal rules for how you engage on twitter?
English