Fwiw, human underperformance on vending-bench is an elicitation problem. The human was only given a set amount of time (iirc 5 hours), whereas the LLMs run until they lose coherence. The human maintained coherence throughout but types slower, and therefore sees less days of simulation.
The human is also the only case with n=1.
There might be a better chance of this becoming widely used if the design is pitched to IKEA (or similar), over being developed in a new, dedicated company
In BE ‘quite’ can also be used sarcastically, becoming a negative intensifier. If you say something’s “quite nice” that could mean it really wasn’t good at all.
Less so once they’ve done enough RLHF
Populations got much bigger post-Industrial revolution, after which very few people were farmers. I’m pretty sure more people who have existed were non-farmers just from that growth, by a huge margin.
But I’m not sure whether or not that should carry over to ancestors. On one hand, you can only have so many ancestors at a time, and explosive industrial population growth doesn’t change that. But smaller farming populations might mean more of my family tree crossing over itself, and so fewer unique farming ancestors?
Update: the cards have all been claimed now (at least for this round)
Thanks for the post!
The Zach Weinersmith quote you mentioned goes further, in a direction that might also be relevant:
″...bravery is not cross contextual… Conversely, recklessness *is* cross contextual. Unsafe sex correlates to drug abuse.”
I’m pretty reckless in some ways, but need to build courage in other areas.Fostering courage where I need it, turning my recklessness into courage where it’s helping me, and minimising it where it isn’t, all seem like worthwhile things to try.
Best guess: because they’ve found the Haskell expertise they were after. That tweet is nearly two years old now, and I’m pretty sure Ed Kmett joined more recently than that.
Is there any data on how people feel like rationality has changed their lives? Somewhat separate to what we’re doing as a community / what some of the most successful rationalists are up to, it seems worth trying to work out how engaging more with rationality changes things for people. I’m pretty sure my engagement with rationality has helped me become stronger; if it turned out not to help most people that would definitely change how I presented things.
Ordered chronologically. In retrospect, I’ve assumed some pretty weak evil forces here, and mostly gone for variations on a needle-in-a-haystack type theme.
Surgically implant the pen in myself
Theseus-duplicate the pen, give one duplicate to the evil forces so they leave me alone
Make a second, cooler-looking pen, ‘accidentally’ lose it to evil forces
Cut a hole in a dictionary/similarly boringish book, put pen in there
bury the pen somewhere random
carve a hole in a tree, put pen in, wait for tree to regrow over it
bury the pen, in a construction site, so it ends up underneath a building (harder to get to)
Hide the pen in the shaft of another, bigger pen
Sew the pen into the lining of a bag
Same but hem instead of lining
Make a box/piece of furniture, hide pen in the wood itself
Take pen apart, hide each bit somewhere different, reassemble as much as I can keep
Ingratiate myself with evil forces, ‘recover’ pen for them, betray them after the fifty years
Hand pen over to evil forces, steal back after 49 years (embed some tracking method in pen that will work in 1905)
Befriend a winemaker, hide pen in corked bottle of wine, keep wine for a long time ‘as an investment’
Hide pen in someone’s grave
Hide pen in someone’s grave, but by hiding it in their body and then waiting for them to die. Hope autopsies aren’t a thing yet.
Take pen somewhere remote, pay someone there to keep it for you
Keep a gun with you ‘to defend against evil forces’. Hide pen in barrel. Hope you never have to actually use the gun.
Convince art collectors the pen is valuable, sell to the highest bidder. Rely on whatever security they use for their collections. Fifty years later, buy the pen back.
Inscribe something on the inside of the barrel (so you can identify it), find as many identical pens as possible, hide them all. Good luck finding the right one in fifty years.
Hide at the bottom of a lake, hope lake doesn’t silt over/get dredged
Use the pen like you would with any other pen. Evil forces assume you’d be hiding the real pen, so that can’t be it.
Taxidermy
Become an expert in art conservation. Re-frame an old, valuable painting (hide pen in frame). Fifty years later, re-frame it again.
Hide pen in the walls of a new building. Demolish after fifty years.
Take one part off pen. Destroy the rest. Attach remaining part to a different pen after fifty years. Claim that if the ship-of-theseus thing before counted, this does too.
Offer pen as collateral for a huge, long-term loan.
Or just open a bank account and put it in your safe-deposit box, I guess.
Convince a member of evil forces to turn their coat. Give pen to them.
Hide pen in fake brick. Make building out of that plus other real bricks. Remember which brick it was, demolish building later.
Bury pen, plant a tree over the top (so that hopefully its root system surrounds it).
Give pen to nomadic group for safe-keeping, hope evil forces can’t find them.
Sell to pawn shop, along with as many other pens as you can get your hands on. Hope evil forces don’t know what kind of pen they’re looking for, or which of the many pawn shops you sold pens to has the right one. Inconspicuously buy back after a while.
Hide pen in walking stick, horribly cripple yourself, claim you just have the walking stick for the obvious reason.
Actually, just hide the pen in someone else’s walking stick instead.
Hide pen in an umbrella handle.
Give pen to church, claim it was, idk, used to sign Jesus’ death warrant. Have them hold on to it as a relic.
Remove ink from pen. Leave in the stationery cupboard at a university. Assume no-one will want it, but no-one will bother to throw it out.
Choose someone at random, give them the pen, tell them to keep it. Make sure evil forces don’t find out who you picked.
Same, but hide it under the floorboards of a random house, without its occupants knowing.
Using future-telling powers, give to a random future classmate of Einstein’s. Tell them to give it to Einstein when they meet him.
Leave pen in the stationery cupboard of the relevant patent office (again without ink if necessary).
Use future telling powers to work out how you’ll manage to hide the pen. Do that.
Decide it doesn’t matter which exact pen Einstein uses. Go into the pen-selling industry anyway. Sell him a different pen when the time comes.
Throw into a river, which probably then silts up and then changes course. Dig up from where the river used to be fifty years later.
Bury pen in an area high in some mineral people might want in fifty year’s time but don’t at the moment (silicon? uranium?). Get in to the mining industry.
Drop the pen in the ocean near the Netherlands. Dig up a tulip farm fifty years later.
Leave the pen in the dead sea for a while, until a huge salt crystal forms around it. Donate to a museum as an interesting geological specimen.
Bury the pen in an already-discovered archaeological site. (Become an archaeologist). Decide there was more to that area than was initially discovered, and dig it up further fifty years later.
I’m glad they don’t have to work....
1. Clothes might include a bobby pin in my hair, which I could use to pick the lock
2. Wait ten years to see if anyone opens the door
3. Get on the wifi, ask someone for help
4. Find other devices in the area, communicate with them (ask for help)
5. (coerce into opening)
6. Post video of myself in room on social media, create outrage campaign
7. (If door is electronically locked, and my phone’s an android) hack the network (and open door)
8. (and hack a bunch of small devices, to ddos the building’s network)
9. (..., to ddos whatever controls the power supply)
10. Work out where I am, get schematic of building – find and cut power from in the room (there’s a light in here right?)
11. Use my phone/shoes to scrape away at the door for ten years, until I wear a hole in it
12. Same but through the floor and tunnel out
13. Same but through a different part of the wall?
14. Same but through the roof? (I’m pretty short though)
15. Hire militia to storm the building
16. Call police
17. Call local army equivalent?
18. Identify and call families of whoever’s locked me up to guilt trip them into letting me out
19. Get on to the UN
20. Call a locksmith
21. Call fire service (they do this kind of thing right?)
22. If I have wifi I might be being monitored. Be really annoying for a few weeks until captors get fed up with me
23. Same but sit and don’t do anything, until they get bored (use phone for entertainment)
24. Use phone/memory to work out if I might be in a(n unusal) simulation. Cloud-compute something expensive from phone to make it harder to run
25. Do something else that might break it (applying usual system-bug-finding approach)
26. Break the fourth wall (in this exercise), walk out through it
27. Find a way to blow up my phone battery near the door
28. (Assuming electronic lock again) Mess with phone electronics to make a taser, use infinite charge to fry the lock
29. Kick the door down (not needing food for 10 years → super strength)
30. Break the lock (similar idea)
31. Punch through the door
32. Pull door off hinges
33. Take apart phone, find something sharp, cut through hinges
34. I keep cards in my phone case – use one to do the lockpicking thing where you swipe it next to the door (this is relatively normal, how did I not think of it earlier?)
35. (Assuming card-swipe lock) just like, swipe my uni id card on the lock and hope I have access?
36. Order pizza to wherever I am, and put in the delivery instructions that to get to me, the delivery person has to open the door. Let them work it out.
37. Hire several hundred mechanical turkers to work out what’s happening
38. Post a babble challenge describing my exact situation on LessWrong, ask everyone for an insane number of ideas, and try the most promising/realistic ones
39. (If previous techy ideas don’t work) Listen to the walls for a pulse width modulator (most rooms have one somewhere), dig it out over the course of several years, use as extra electronics for previous ideas.
40. Same thing, but this also gives me access to the building’s electronics. Physically wiretap the building’s network, repeat previous hacking-style approaches
41. Same, but just put a stupid voltage across it (using infinitely charged phone) and see what happens
42. Same, but it’s probably in between the walls in a big hole. Escape through that.
43. Similar idea: dig through wall until you get to insulation (must be some if cold isn’t a problem for ten years) – which will be easier to dig through, and might lead into the roof or somewhere easier to get out from (probs not great for respiratory health though – use shirt as a face mask)
44. Pull a nail out of the walls and bash into the lock? (/hinge/other single point of failure in the door)
45. Wait out the ten years, learning a bunch of new stuff while I wait. The person who was originally locked in the room no longer exists there, and so has kind of escaped.
46. Similar thing but kill myself? (Kind of dark, not recommended)
47. Just going to sleep would do the same thing under some philosophies of self?
48. I suppose crudely lobotomising myself would also work here
49. Similarly, ten years’ worth of solitary, internet-aided meditation would probably erode my sense of self enough for the same effect.
50. Read a lot of really good fiction – it’s a mental escape, if nothing else
Nice post! I have one thing to add about timing. Because Vitamin D is fat-soluble, there’s slightly better uptake if you take it with a relatively fatty meal or with milk. I’m not sure how much of a difference this makes though; taking it in the morning (with a not-fatty meal) is still enough to fix my deficiency.
Weird question, but if your marker action is doing something with your hands, what do you do if you’re holding something at the time?
2 cents, not based on source code or these specific claims: It’s a terminal app written in javascript, which is often slow/flickery because it redraws the entire screen whenever something changes. Whereas Codex fast-followed & very quickly was rewritten in Rust, and its ux is noticeably snappier.